Mining DD

What they build

Realm builds an AI operations layer for mines that plugs into infrastructure the site already runs (CCTV, SCADA, sensors, drones, data lakes) via read-only integrations, then runs computer vision and sensor fusion on top. The product is now clearly productized into seven named use cases: unsafe-behavior detection, PPE-compliance enforcement, froth-flotation optimization (computer vision on tank flow and bubble characteristics to recommend reagent adjustments), a longwall-shearer wear/performance assistant, an MSHA citation assistant that mines regulatory documents for violation patterns and root causes, 24/7 SCADA alerting on fan health and atmospheric safety, and automated pre-shift safety briefings. What is genuinely hard: the vision models are trained on real underground footage under dust, low light and water, and the system is designed to fuse video with sensor data to suppress false positives and adapt per site. That is a forward-deployed, integration-heavy wedge, not a generic AI dashboard. Stated deployment cadence is aggressive: week 1 integration and model training, week 2 live, week 4 pattern analysis, with early wins claimed inside 30 days. Current stage is live-in-production at active mine sites.

Team

Two unusually senior founders, both Palantir alumni. Eugene Marinelli (co-founder) was an early Palantir engineer (2011-2016) and then founder/CTO of Blend, a now-public mortgage-tech company he scaled for roughly a decade; he is the technical anchor. Myles Recny (co-founder) was a Palantir strategist (2015-2019) and co-founded Even, a financial-wellness platform acquired by Walmart; he is the GTM/BD anchor. The prior dossier listed a junior third name (Dylan Barnacle, ex-Palantir FDE-track); that remains a plausible early deployment hire rather than a founder. Team size reads small and lean: sources cluster around 7-11 people, headquartered in Austin TX with a New York and UT-Austin footprint. Hiring is real but modest, spanning software engineering (incl. new-grad), business development and strategic partnerships. This is a classic Palantir-mafia forward-deployed team profile.

Funding

New this pass: the investor syndicate is now identified even though the amount stays undisclosed. Backers include 8VC, Alt Capital (Jack Altman's fund), Conversion Capital, Narya (the Peter-Thiel-linked fund), and Polymath Capital Partners, six investors in total. That is a materially stronger read than the prior confirmed-stealth dossier, which showed $0 and could only infer a pre-seed. This looks like a seed round from a name-brand syndicate rather than a friends-and-family pre-seed. Exact amount and valuation remain behind Crunchbase/PitchBook paywalls, but a seed of low-to-mid single-digit millions is the reasonable estimate given the syndicate and headcount. With a sub-15-person team, runway is likely comfortable (18-30 months) even at the low end. Almost certainly under any reasonable cap.

Traction

VERIFIED: a named reference customer, Oak Grove Mining, with testimonials attributed to CEO Mans Larsson and Manager of Special Projects Matt Martin on Realm's own site, plus the explicit claim that everything built ships to active sites and runs in production. CLAIMED (single-customer, self-reported): a 25% reduction in unplanned downtime and early wins within 30 days. The MSHA-citation and compliance angle is a smart, verifiable-value hook in a heavily regulated US mining context. Weaknesses in the evidence: only one customer is publicly named, and the headline metrics are not independently audited. Active hiring across BD/partnerships suggests they are in an early land-and-expand phase rather than at scale.

Market and competition

The niche is AI safety, compliance and plant-optimization software for mine operators, an underserved, conservative, regulation-heavy industry. TAM sanity: global mining spends tens of billions annually on safety, compliance and lost-production downtime, so even a thin software slice is a real multi-hundred-million to low-billions software opportunity. Competitors/adjacent: MaxwellGeoSystems and Newtrax/Sandvik (underground monitoring), Motion Metrics (Caterpillar-owned, vision for shovels/trucks), Plotlogic (ore-characterization vision), Intellisense.io (plant optimization AI), plus in-house data teams at the majors and generic industrial-vision vendors. Realm differentiates on breadth (one platform spanning safety + compliance + plant optimization), read-only fast integration, and a Palantir-style forward-deployed model. Incumbency dynamics favor whoever can integrate into legacy SCADA and win operator trust; the majors move slowly and prefer proven references, which is both the moat and the sales barrier.

Bull case

• Elite, proven founders (public-company CTO in Marinelli, Walmart-acquisition operator in Recny) attacking a genuinely underserved industry with a forward-deployed Palantir playbook.
• Now-verified brand-name syndicate (8VC, Narya, Alt Capital, Conversion, Polymath) de-risks the confirmed-stealth funding uncertainty from the prior dossier.
• Productized into seven concrete use cases already shipping to a live site, with a fast 30-day deploy motion and a real compliance hook (MSHA) that maps to hard-dollar regulatory pain.
• Broad single-platform surface (safety + compliance + optimization) supports land-and-expand within each mine.

Bear case

• Only one customer (Oak Grove) is publicly named; the traction and the 25% downtime metric are self-reported and unaudited.
• Mining sales cycles are long, reference-driven and conservative; a lean sub-15 team may struggle to cover enterprise BD across many sites.
• Breadth risk: seven use cases at a tiny team can mean none is deep enough to be a must-have versus focused point competitors.
• Funding amount and valuation still undisclosed, so pricing/dilution and true runway remain inferred.

Key risks

• Concentration and proof risk: failure to convert beyond a single named customer into repeatable multi-mine deployments would stall the whole thesis.
• Enterprise sales drag in mining outrunning a lean team's capacity, letting an incumbent (Caterpillar/Motion Metrics, Sandvik/Newtrax) bundle a good-enough version.
• Model reliability in harsh underground conditions: false positives or a missed hazard could destroy operator trust and the safety-value narrative.

Questions for the founder

1. Beyond Oak Grove, how many active paying sites are live, and what is annualized recurring revenue versus pilots?
2. What exactly was the seed round size, valuation and current runway, and who led?
3. Of the seven use cases, which one is the wedge that customers pay for first, and what is its measured, third-party-verifiable ROI?
4. What is the integration time and failure rate against legacy or air-gapped SCADA environments at real mines?
5. How do you defend against Caterpillar/Motion Metrics or Sandvik bundling comparable vision/safety features into equipment they already sell into these mines?

What they build

Sisir Radar is an India-based deep-tech company designing high-resolution Synthetic Aperture Radar in L- and P-bands, plus ground-penetrating radar (GPR) and drone-borne hyperspectral imaging, for all-weather, day-night, foliage-penetrating situational awareness. The flagship program is India's first privately built L-band SAR satellite, with a claimed 0.75m L-band and 2.5m P-band resolution described as roughly 10x better than prevailing standards, plus a large (6m-class) unfurlable antenna. New this pass on timeline: the company and press target 2026 for launch, but the satellite-platform partner XDLINX states a 2027 launch, so the realistic launch window is 2026-2027, not a hard 2026. What is genuinely hard: L-band spaceborne SAR with a large deployable antenna is elite RF/antenna and space-systems engineering, and Sisir also has a shipping terrestrial line, the LMD-GPRXI-INDRA handheld GPR, extending the radar IP into IED detection, infrastructure monitoring, mineral exploration and archaeology.

Team

The load-bearing founder is Dr. Tapan Misra, an ex-ISRO scientist and former ISRO/SAC director who worked on India's first SAR satellite mission (RISAT), a genuinely elite SAR/space pedigree; this is correctly the claim Sisir should be judged on. New this pass: co-founders now surface as Soumya Misra and Urmi Bhambhani (the latter noted as an ex-Summachar co-founder), rounding out the founding team beyond Tapan alone. The prior dossier's Apple-COO name (Shaibal Ghosh) remains misattributed dataset noise and is not the technical founder. Team size clusters around 40-43 with roughly 17 engineers, a strong technical density for a hardware-heavy deep-tech at this capital level. XDLINX Space Labs (CEO Rupesh Gandupalli) is the satellite-platform integration partner, not staff.

Funding

Two rounds, ~$8.5M cumulative. A ~$1.5M seed (Feb 2025) and a ~$7M (INR 63.5 crore) Series A led by 360 ONE Asset with existing investor Shastra VC participating, announced Dec 15, 2025. Under a $10M cap but close to it. Stated use of funds: satellite design and manufacturing in India, expanding R&D/engineering teams, and moving from prototypes to full orbital missions. The obvious tension: $8.5M is thin to design, build and launch a large-antenna L-band SAR satellite; substantial follow-on capital will be needed and near-term. Non-equity leverage exists via defense contracts (see traction).

Traction

VERIFIED and unusually strong for the stage: Sisir won two iDEX (Innovation for Defence Excellence) challenges to build specialized SAR payloads for the Indian Air Force, including an iDEX contract for an L- and P-band SAR payload and a 6m-diameter unfurlable antenna, awarded on stage by the Defence Minister at DefConnect 2024. That is a real government customer and non-dilutive validation. Additional VERIFIED signals: a formal launch/integration partnership with XDLINX Space Labs, and a shipping handheld GPR product (LMD-GPRXI-INDRA). CLAIMED/less specific: broader defense and industrial partners for surveillance and detection, and mining/agriculture/environmental applications for the satellite. Revenue figures are not disclosed, but the iDEX defense contracts are the strongest verified traction of the four companies here.

Market and competition

The niche is high-resolution, dual-use L/P-band SAR (space and terrestrial) for defense, disaster management, agriculture, mining and infrastructure, anchored by India's sovereign-capability and import-substitution push. TAM sanity: the global commercial SAR/Earth-observation market is multi-billion and growing, and Sisir additionally taps Indian defense procurement, a large captive buyer. Competitors: on satellites, ICEYE and Capella Space (X-band, so Sisir's L/P-band foliage-penetration is a genuine differentiator), plus Umbra and India's own SAR startups (e.g. GalaxEye) and ISRO itself; on GPR/terrestrial, standard geophysical-instrument vendors. Sisir differentiates on band choice (L/P-band penetration vs the X-band crowd), an ISRO-pedigree founder, and a sovereign-India defense wedge. Incumbency dynamics: ISRO relationships and iDEX are both moat and dependency.

Bull case

• Elite, correctly load-bearing founder (Tapan Misra, ex-ISRO, RISAT SAR pioneer) plus ~17-engineer technical density: rare depth for the capital raised.
• Verified, non-dilutive defense traction: two iDEX wins and a Defence-Minister-awarded IAF contract for the L/P-band payload and 6m unfurlable antenna.
• Differentiated band choice (L/P-band foliage penetration) versus the X-band incumbents ICEYE/Capella, plus a real launch partner in XDLINX.
• A shipping terrestrial GPR product diversifies revenue beyond the long satellite timeline.

Bear case

• $8.5M is brutally thin to actually build and launch a large-antenna L-band SAR satellite; heavy near-term dilution or delay is likely.
• Launch timeline is soft: 2026 per the company/press but 2027 per the platform partner XDLINX, so slippage risk is already visible.
• India-based dual-use/defense deep-tech carries geographic, regulatory and export-control complexity for some investors.
• Heavy dependence on Indian government/iDEX procurement concentrates customer and political risk.

Key risks

• Capital risk: inability to raise the much larger sums needed to reach orbit before running down the ~$8.5M base.
• Launch/execution risk: schedule slip on a hard large-deployable-antenna L-band satellite (2026 vs 2027 already in tension).
• Customer-concentration risk: over-reliance on Indian defense (iDEX/IAF) if commercial and export demand is slower to materialize.

Questions for the founder

1. What is the fully-loaded cost and financing plan to actually launch the L-band SAR satellite, and how much more capital is required beyond the ~$8.5M raised?
2. Is the launch 2026 or 2027, and what are the specific gating milestones (antenna qualification, payload integration with XDLINX)?
3. What is current revenue split between the iDEX/IAF defense contracts and the GPR product line?
4. How defensible is the L/P-band advantage versus ICEYE/Capella/Umbra as they add constellations and revisit?
5. What export-control and dual-use constraints affect selling SAR data or hardware to non-Indian customers?

What they build

ExploreTech is software-driven critical-mineral exploration that fuses probabilistic geophysical inversion with AI-powered drill-target planning to build 3D subsurface models and optimize drilling in real time, running on high-performance CPU/GPU/MPI compute. The productized layer is two cloud APIs: Inverter (geophysics/probabilistic inversion) and Driller (drill-target optimization/planning). The genuinely hard part is the probabilistic inverse theory itself, which is exactly the founders' PhD research, and translating uncertainty-quantified geophysics into concrete drill recommendations a mining geologist will act on. New strategic direction this pass: ExploreTech is deliberately moving beyond a pure software/service vendor toward extreme vertical integration, i.e. taking mining claims and using its own tech to discover deposits, which changes the risk/return profile from SaaS toward explorer economics. Stated goal: compress discovery timelines from 10-plus years to 1-2 years and hit the same drilling results with roughly 85% fewer boreholes.

Team

Two verified Stanford PhD co-founders with textbook founder-market fit, both from mining backgrounds who met as first-year PhD students in 2017. Alex Miltenberger (PhD, Stanford Geophysics; dissertation on probabilistic inverse theory under Prof. Tapan Mukerji) is the geophysics/inversion brain. Tyler Hall (PhD, Stanford Geological Sciences; TomKat Graduate Fellow working on AI subsurface-exploration planning) is the geology/planning brain. The company was founded in 2023 and is based in San Diego. It is very small, roughly 7 people with a high (~0.43) engineering fraction. The founders describe culture as open communication, alignment and intensity. This is deep, verified domain expertise commercializing the founders' own research.

Funding

New this pass: the round is now fully specified. A $2M pre-seed closed May 29, 2025, led by Primary Venture Partners, with participation from KDX Ventures and Climate Capital plus unnamed angels; this was the first external round. Earlier the company was bootstrapped for about a year on service revenue, and it holds a Stanford TomKat Center Innovation Transfer Grant (Sept 2023). So total external equity is about $2M plus grant, comfortably under cap. Use of funds: extend field data collection and move from computing services toward the vertically integrated exploration model. At ~7 people, $2M gives a meaningful runway (likely ~18-24 months), but the vertical-integration pivot (owning claims, funding drilling) is capital-hungry and will pressure that runway or require a larger raise.

Traction

The strongest verified traction of the four on pure prediction performance. VERIFIED: at Giant Mining's Majuba Hill copper-silver-gold project in Nevada, ExploreTech's models predicted a mineralized zone; hole MHB-36 intersected chalcopyrite (copper) mineralization starting at ~650 ft, within ~50 ft of the predicted 600-700 ft window, publicly disclosed by the operator. The operator states ExploreTech's drilling recommendations have correctly intersected the source of geophysical anomalies 7 out of 7 times, the first public disclosure of that validation record. Also VERIFIED: named mining partners/collaborations (Giant Mining, Reyna Silver at Guigui, Prismo Metals, Honey Badger Silver) and, per the founders, a bootstrapped period serving ~15 exploration companies as a tech-enabled service provider. Strong conference/press presence (AEMA annual meeting, PDAC, S&P Global, Northern Miner, The Intelligent Miner). This is real, third-party-disclosed, in-the-ground validation, not just a pitch metric.

Market and competition

The niche is AI/geophysics exploration-targeting software (and now optionally an integrated explorer) for critical minerals: copper, lithium, nickel, rare earths. TAM sanity: exploration is a large slice of the ~12-13 billion dollar annual global exploration spend, and if ExploreTech captures deposit value directly via claims the upside is explorer-scale rather than SaaS-scale. Competitors: KoBold Metals (the giant, AI-driven exploration, far better funded), Earth AI (AI targeting plus its own drilling), VerAI, GeologicAI, and classic geophysical-software incumbents (Seequent/Leapfrog, Geosoft). ExploreTech differentiates on rigorous probabilistic inversion (uncertainty-quantified, PhD-grade) and a validated in-the-ground hit record, but it is small and capital-light versus KoBold and Earth AI. Incumbency dynamics: conservative junior/major miners buy on references and de-risking, which favors ExploreTech's growing public validation but lengthens the sales cycle.

Bull case

• Best verified traction of the group: a 7-for-7 anomaly-source hit record and a publicly operator-disclosed Majuba Hill copper intercept within ~50 ft of prediction.
• Two Stanford PhDs commercializing exactly their own research (probabilistic inversion + AI drill planning), an unusually clean founder-market fit.
• Fully specified, under-cap funding (Primary-led $2M pre-seed plus Stanford grant) after a disciplined bootstrapped period serving ~15 companies.
• Optional vertical-integration path (owning claims) converts a modest SaaS opportunity into explorer-scale upside if the targeting edge is real.

Bear case

• Tiny (~7 people) and capital-light ($2M) against a KoBold Metals that has raised hundreds of millions for the same thesis.
• The vertical-integration pivot toward owning claims is capital-hungry and dilutive to focus, and pushes the company into cyclical explorer economics.
• Validation, while real, is still small-n and disclosed by the mining partners themselves; long-run hit rate on harder targets is unproven.
• Conservative junior/major mining buyers mean long, references-driven sales cycles that a 7-person team must grind through.

Key risks

• Scale/capital risk: being out-resourced by KoBold/Earth AI before the validation record compounds into durable commercial traction.
• Strategy risk: the software-to-vertically-integrated-explorer pivot could over-extend a tiny team and burn the modest $2M on drilling/claims.
• Statistical risk: a public miss after the 7/7 streak (small sample) would disproportionately dent the credibility the whole thesis rests on.

Questions for the founder

1. How many of the 7/7 validated holes were blind pre-registered predictions versus post-hoc fits, and across how many distinct projects and deposit types?
2. Is the business ultimately software/SaaS, tech-enabled service, or an integrated explorer owning claims, and how does that decision drive capital needs?
3. What is current revenue: recurring API/software, per-project service fees, or claim/royalty upside, and how much is contracted?
4. How do you compete for the best exploration targets and talent against KoBold's and Earth AI's much larger war chests?
5. What is runway on the $2M given the vertical-integration ambition, and what is the size/timing of the next raise?

What they build

Durin builds remotely operated, semi-autonomous core-drilling rigs for mineral exploration, attacking a category of equipment its founder argues has barely changed since the 1950s. The first rig is a manually operated but heavily sensor-instrumented core rig capable of boring roughly 300 meters deep at a 2.5-inch diameter; the sensors stream real-time mineralogical and structural data to geologists (mineralogy in minutes, assay data in days not months). The hard part is twofold: building rugged drilling hardware at all, and collecting enough instrumented real-world drilling data to train autonomy models that can handle downhole variability. Strategy is a four-stage roadmap: BUILD safer smarter rigs, OPERATE them on active North American exploration projects, SCALE production for global deployment, then CONSOLIDATE by vertically integrating geological modeling, lab testing and drillhole planning. Full unattended autonomy is targeted in 2-3 years, with humans still needed for supply, monitoring and sample retrieval. Currently pre-commercial, running pilot drilling in Nevada.

Team

New this pass: the team is bigger than the founder-only picture the prior dossier implied. Beyond founder/CEO Ted Feldmann (third-generation mining professional, BSc Georgia Tech, the domain-native founder-market fit), Durin now shows a COO, David Venegas (operations/engineering, running day-to-day and tech integration), and a Head of Geoscience, Christopher J. Seligman (certified professional geologist leading geological R&D and data integration). The prior dossier's ex-SpaceX early engineer (Joshua Schertz) fits the ~22-person, engineering-dense profile. So the team is founder-domain-fit plus operational and geoscience leadership plus SpaceX-caliber engineering, a credible hardware-startup spine. Hiring skews engineering and field operations.

Funding

$3.4M pre-seed, April 2025, led by 8090 Industries, with a strong deep-tech syndicate: Andreessen Horowitz, Lux Capital, 1517 Fund, Bedrock, Champion Hill, Contrary, and Day One Ventures. This is a16z-verified and well under any cap. No Series A has been announced as of mid-2026. For a hardware/robotics company burning on rig fabrication, field crews and a ~22-person team, $3.4M is thin; a follow-on raise is likely needed within 12-18 months of the pre-seed, so a 2026 raise should be expected even though none is yet public. Business model reads as drilling-as-a-service (operate rigs on projects and earn drilling revenue, reinvesting in R&D and data), eventually vertically integrated, rather than selling rigs.

Traction

VERIFIED: pilot drilling underway in Nevada and the pre-seed syndicate. CLAIMED/EARLY: no named paying customers or contracted drill programs are public yet; revenue signals are absent. The founder's market framing is the strongest evidence of thesis validity rather than of traction: about 70% of exploration capital goes to drilling, labor is roughly 60% of drilling contractors' costs, the US has an acute shortage of qualified drillers, only about 3 of 1,000 drill attempts find a deposit, and global exploration spend was about 12-13 billion dollars in 2023. Plan to expand into South America by end of 2026. So traction today is a working first rig and active pilots, not booked revenue.

Market and competition

The niche is modern, instrumented, eventually-autonomous exploration core-drilling for critical minerals, tied to US/Western reindustrialization and supply-security narratives. TAM sanity: exploration drilling is a meaningful slice of the ~12-13 billion dollar annual exploration spend, plus the broader drilling-services market, so the addressable pool is real and multi-billion. Competitors: legacy drilling-services incumbents Boart Longyear (150 years, moving toward hands-free rod handling since 2016), Epiroc (Pit Viper autonomous, but open-pit production not slim-hole exploration) and Sandvik (AutoMine); and a directly comparable startup, XtremeX Mining Technology, whose automated hybrid-power exploration rig is running a funded field trial with Ivanhoe Electric at Santa Cruz in Arizona to 2,000m. Durin differentiates on a data-first, autonomy-from-the-ground-up design and a domain-native founder, but XtremeX shows it is not alone and the incumbents own the customer relationships and field service networks.

Bull case

• Genuine founder-market fit (third-generation miner) now backed by a real operational and geoscience bench (COO, Head of Geoscience) and SpaceX-caliber engineering.
• Elite deep-tech syndicate (a16z, Lux, 8090, Bedrock, 1517) validating a supply-security thesis with strong tailwinds around critical minerals and reindustrialization.
• Attacks the single largest cost line in exploration (drilling, ~70% of capital) where labor shortage makes automation economically compelling.
• Working first rig plus active Nevada pilots means real hardware in the ground, not slideware, and a data flywheel toward autonomy.

Bear case

• $3.4M is very thin for hardware/robotics; the company is likely dependent on a near-term follow-on raise it has not yet announced.
• No named paying customers or contracted drill programs are public; traction is pilots, not revenue.
• Full autonomy is 2-3 years out and downhole autonomy is genuinely hard; timeline slippage is the base-rate outcome in drilling automation.
• A directly comparable competitor (XtremeX) is already field-trialing with a real miner (Ivanhoe) to greater depth, and incumbents own the service relationships.

Key risks

• Capital risk: running out of runway before proving drilling-as-a-service unit economics, in a capital-hungry hardware business that VCs fund cautiously.
• Autonomy timeline risk: if unattended operation slips well past 2-3 years, the differentiation collapses to just another instrumented manual rig.
• Commercial-adoption risk: conservative junior/major miners may not switch drilling contractors to an unproven startup without a long reference trail.

Questions for the founder

1. What are the current pilot economics: cost per meter versus a conventional contractor, and is anyone paying for the Nevada drilling yet?
2. What is runway on the $3.4M, and what is the size/timing of the next raise you are planning?
3. How far along is the autonomy stack today: what is actually automated versus remotely operated, and what is the honest milestone map to unattended operation?
4. How do you win against XtremeX (already trialing with Ivanhoe to 2,000m) and against incumbents' field-service networks?
5. Is the long-term model drilling-as-a-service, selling rigs, or the full vertically integrated explorer, and how does that change capital intensity?

What they build

VOLUNA builds Gammine, an autonomous field-deployable exploration system that uses neutron-induced nuclear reactions (neutron activation analysis) combined with gamma spectroscopy to measure elemental concentrations of the ground directly in the field, in hours, with no lab and no sample shipping. It miniaturizes nuclear sensors onto autonomous ground rovers, with an aerial drone version (Condor) planned, and sells it as exploration-as-a-service rather than as hardware. The physics is the whole company: fire neutrons into rock, read the characteristic gamma signatures of the elements that get activated, and map grade in real time. If it works at useful penetration depth and throughput it collapses the months-long sample-assay-lab loop into same-day answers, which is a genuinely different capability from every AI-on-existing-data competitor in this cohort. The two unsolved hard problems remain subsurface penetration depth and the regulatory path for mobilizing neutron sources across borders and terrain.

Team

The deep-tech pedigree is now fully verified rather than merely claimed. CEO and cofounder Alexander Strange worked at Blue Origin on structural design for the New Glenn launch vehicle, then at Astranis on high-orbit internet satellites; he holds a Harvard MBA, a Stanford MS in Aero/Astro, and a BS in Engineering Mechanics and Astronautics from UW-Madison (confirmed via LinkedIn and multiple profiles). CTO Dr. Aaron Olson is a former NASA Kennedy Space Center physicist (Swamp Works, lunar in-situ resource utilization and neutron sources). COO Elias Fernandini is a fifth-generation Peruvian miner with a Harvard Kennedy School background, supplying both mining domain access and the Peru pilot channel. Robert Frady adds deep laboratory-instrumentation engineering. The most important new team fact is external: VOLUNA's strategic partner Clandestine Materials Detection is led by Gerald Kulcinski, Grainger Professor of Nuclear Engineering emeritus and Director of the UW-Madison Fusion Technology Institute, a National Academy of Engineering member and two-time NASA medalist whose lab is literally building a fusion neutron generator compact enough to fly on a drone. That is arguably the single most credible neutron-source pedigree available anywhere, and it de-risks the core physics and hardware bet materially.

Funding

The raise is deliberately undisclosed and sits at pre-seed/seed scale. Named backers are Pillar VC (invested September 2025), 2045 Ventures, Breakthrough Energy Fellows (Cohort 5), Switch, MBA Ventures, plus selection into the Founders Factory x Rio Tinto mining accelerator from 500-plus applicants. PitchBook and Crunchbase-adjacent sources confirm a pre-seed/seed round in 2025 with roughly nine investors but decline to publish the dollar figure; no valuation is public. Best estimate remains low single-digit millions or less, definitively under any reasonable cap. Note the corrected headquarters: VOLUNA is based in Madison, Wisconsin (co-located with CMD and rooted in the UW-Madison nuclear/fusion ecosystem), not Boston, which also explains the CMD tie-up. The prior wrong-entity trap stands: do not attribute Terra AI's Khosla/BHP-backed ~20M+ round to VOLUNA; they are different companies.

Traction

The Peru pilot is real but the counterparty is still not publicly named. What is confirmed: VOLUNA built its first Gammine unit and deployed it to a copper project in Peru in fall 2025, and after one week in the field it was delivering same-day elemental data with no labs and no delays. The undisclosed counterparty is almost certainly reached through COO Elias Fernandini's fifth-generation Peruvian mining family network, which is the likeliest access channel, but VOLUNA has not disclosed the customer, whether money changed hands, or the depth and accuracy achieved. The March 2026 CMD strategic collaboration is the more materially verifiable event: co-developed IP, CMD contributing decades of neutron-detection capability, regulated facilities, and specialized expertise, with economic reward to CMD shareholders tied to VOLUNA's success and Wisconsin Alumni Research Foundation holding a stake in CMD. So the strongest proof points are pedigree and partnership, not yet a signed paying customer or demonstrated production-grade depth performance.

Market and competition

VOLUNA is the odd one out and the most differentiated of the four: it is a hardware-plus-physics measurement company, while MinersAI, Mineflow and BiaTech are all software-on-existing-data plays. That means VOLUNA is not really competing with them for the same budget; it competes with the assay lab and the drill program itself, and complements rather than replaces AI-target-generation tools (you still need something to decide where to send the rover). Against the funded field it is closest in ambition to KoBold Metals in that both attack the discovery cost curve, but VOLUNA does it with a novel sensor rather than data fusion, and at a fraction of KoBold's ~537M war chest. Its true peers are other in-field spectrometry and geochemical-sensing startups, a much thinner competitive set than the crowded AI-data layer. If the neutron physics works at depth, VOLUNA owns a capability none of the software players can replicate; if it does not, there is no software pivot to fall back on.

Bull case

• Fully verified, elite deep-tech team (Blue Origin, Astranis, NASA Kennedy) plus a National-Academy-of-Engineering neutron physicist via the CMD partnership, an almost uniquely credible bench for this exact hard problem.
• The most genuinely differentiated technology in the cohort: in-field, same-day elemental mapping competes with the lab, not with crowded AI-data tools, and has no direct software substitute.
• A live copper pilot in Peru already delivering same-day data within a week, plus Peru mining-family access via the COO for a warm channel.
• Selective validation (Breakthrough Energy Fellows, Rio Tinto accelerator from 500-plus applicants, Pillar VC) and a partner whose lab is literally building drone-flyable neutron sources.

Bear case

• Genuinely pre-product: one alpha pilot, no named paying customer, no disclosed depth or accuracy results, no revenue, and no patents found.
• Core physics risk unresolved: useful subsurface penetration depth and throughput for mobile neutron activation is unproven at commercial scale.
• Heavy regulatory overhang and dependence on partner (CMD) IP and licensed facilities to mobilize neutron sources across jurisdictions.
• Hardware capital intensity and long sales cycles into conservative miners; no software pivot if the sensor thesis fails.

Key risks

• Neutron penetration depth may prove too shallow for the deposits that matter, limiting the addressable use case to near-surface targets.
• Regulatory approval for mobile neutron sources could be slow, jurisdiction-specific, and a permanent friction on deployment.
• Dependence on CMD for critical IP and facilities creates counterparty and licensing risk.
• Capital intensity of hardware plus long mining sales cycles could outrun a modest pre-seed runway.
• Key-person concentration in a small, elite but tiny founding team.

Questions for the founder

1. Who is the Peru copper counterparty, was it a paid engagement, and what penetration depth and grade accuracy did Gammine actually achieve in the field?
2. What is the exact pre-seed amount, valuation, and runway, and how much is grant/fellowship versus priced equity?
3. What is the regulatory pathway for deploying mobile neutron sources commercially, and in which jurisdictions are you cleared or seeking clearance?
4. How does the CMD IP arrangement work in detail: what do you own versus license, and what happens to VOLUNA if that relationship ends?
5. At what deposit depth does neutron activation stop being useful, and how large is the addressable market above that limit?

What they build

MinersAI builds a cloud data-and-analytics platform that ingests messy exploration geoscience (geological, geochemical, geophysical) from historical and current projects, cleans and standardizes it into an AI-ready form, and layers digital-prospecting tools on top: advanced search, filtering, pattern recognition, geochemical analysis, and mineral-probability or target-generation mapping for critical minerals like lithium, nickel, copper, cobalt and rare earths. It also runs a geodata marketplace. The company positions itself as a computational discovery partner to in-house geoscience teams rather than a replacement, which is the right framing for a conservative industry that will not hand its geology to a black box. This is a data-infrastructure and workflow play more than a proprietary-discovery play: MinersAI does not own mineral rights or drill, so its moat is the standardized data corpus and the target-generation models trained on it, not exclusive access to the ground. Founded July 2023, Boulder Colorado plus Brussels, and as of May 2026 a third node in Perth.

Team

The 2-of-18 engineering worry from the prior dossier does not survive a look at the current team page. CEO and cofounder Mason Dykstra (PhD geologist, active critical-minerals voice) and CPO and cofounder Tomi (Thomas-Louis de Lophem) lead. On the build side there is a genuine bench: Brenden as Head of Engineering, Hubert on full-stack, Ahmed as ML and Data Engineer, plus Tyler and Christian as ML Geodatascientists, Molly as Geodata Engineer, and Simon as GeodataScientist, with Ken as Project Lead. That is six to seven people who actually ship software and models, not two. The rest are geoscience (Abbie on data models, Fernando and Jose as geologists). So the composition is a roughly even split of builders and domain scientists, which is arguably ideal for this problem. The newest hire is the most commercially significant: Leon Morgan, appointed May 2026 as Global Chief Revenue Officer and APAC lead out of Perth, an Australian mining-sales hire meant to convert the platform into revenue. Individual employment histories are still not exhaustively verified, but the shape of the team is credible and better-engineered than the tracker headcount implied.

Funding

The load-bearing new fact is that BHP Invent is confirmed as a direct named investor on MinersAI's cap table (per Crunchbase-sourced investor lists), not merely an accelerator relationship. The syndicate reads BHP Invent, Speedinvest (partner Andreas Schwarzenbrunner), Creative Destruction Lab, Wilbe Capital, Mission1Capital, We Love Founders, plus Techstars Sustainability Paris and Slush affiliations. Total raised is low seven figures: trackers cluster around 930K to 1.22M across what is described as a single pre-seed dated early February 2024, with outlier readings as low as 120K. No disclosed valuation. Critically, no public evidence shows a commercial supply or offtake agreement attached to the BHP investment. It reads as a strategic corporate-venture pre-seed check giving BHP optionality and a window into the tooling, not a locked commercial contract. That distinction matters: BHP Invent money is a strong validation signal but does not by itself equal BHP revenue.

Traction

MinersAI's flagship named pilot is with Chilean Cobalt Corp (OTC: COBA) at the La Cobaltera cobalt-copper project in Chile's San Juan district, announced 2025 with technical milestones targeted for H2 2025. The important nuance the prior dossier lacked: this is a triple AI partnership, MinersAI alongside Chile-based Mineral Forecast and US-based TerraSpace, so MinersAI is one of three vendors on a single junior's project, not the sole or exclusive provider. No financial terms disclosed; a pilot with a micro-cap junior is validation of the workflow, not proof of paying enterprise revenue. The May 2026 Perth HQ and CRO hire, with a follow-the-sun 24/7 support model across Australia, Europe and North America, signal a genuine commercial push into the largest exploration market on earth. Still, as of now there are no disclosed marquee paying customers and no revenue figure. Traction is real but early: accelerator-vetted, one named junior pilot shared with competitors, and a fresh sales apparatus not yet shown to convert.

Market and competition

Head to head with the other three in this cohort, MinersAI is the pure software-and-data-infrastructure play, closest in category to Mineflow but far broader (full data standardization plus marketplace versus Mineflow's narrow deposit-shape prediction). Against the funded giants it sits well below the discovery-and-own model: KoBold Metals (~537M raised, ~3B valuation, actually finding and developing deposits like the Zambian copper find), GeologicAI (~44M Series B, Canadian, core-scanning hardware plus AI), and Earth AI (~20M Series B, drills its own targets). MinersAI is capital-light and asset-light by comparison, which caps both burn and upside: it sells picks and shovels to explorers rather than betting the balance sheet on the ground. Its defensible edge is the BHP Invent relationship and a standardized multi-project geodata corpus that compounds with every dataset ingested. Its vulnerability is that target-generation and data-cleaning tooling is a crowded, commoditizing layer, as the Chilean Cobalt triple-vendor pilot itself demonstrates.

Bull case

• BHP Invent as a direct cap-table investor is a rare strategic validation from the world's largest miner and a credible path to a marquee commercial anchor and follow-on capital.
• Team is better-engineered than trackers suggested: six to seven genuine builders plus a strong geoscience bench and PhD-geologist CEO, a rare balance for mining tech.
• Capital-light, asset-light model means low burn and low downside; a low-seven-figure raise buys a long runway and multiple shots on goal.
• May 2026 Perth expansion with a seasoned mining CRO puts the company inside the world's deepest exploration market with a real sales apparatus.

Bear case

• No disclosed paying customers or revenue; the one named pilot (Chilean Cobalt) is with a micro-cap junior and shared with two competing vendors.
• Data-standardization and target-generation tooling is a commoditizing layer with many entrants; the moat may be thinner than the marketplace narrative implies.
• Asset-light model caps upside; unlike KoBold or Earth AI, MinersAI captures none of the discovery value it helps create.
• BHP investment carries no visible attached commercial agreement, so the strategic-backer story may not convert to strategic revenue.

Key risks

• Commoditization of AI target-generation compressing pricing and differentiation.
• Dependence on junior-explorer budgets that evaporate in a commodity downcycle.
• Key-person concentration in the two cofounders for domain credibility and fundraising.
• Strategic-investor optionality (BHP) could become a ceiling if it deters competing majors from adopting the platform.
• Under-capitalization relative to funded rivals if the market shifts to capital-intensive discover-and-own models.

Questions for the founder

1. What exactly does the BHP Invent relationship include beyond the equity check, and is any commercial pilot, data-sharing, or offtake agreement attached?
2. How many paying customers and what ARR does the platform have today, versus unpaid pilots and accelerator engagements?
3. In the Chilean Cobalt triple-vendor pilot, what is MinersAI's specific deliverable and how do you avoid being commoditized against Mineral Forecast and TerraSpace?
4. What is the true current headcount, the full-time versus advisor split, and the burn rate on a roughly one-million-dollar raise?
5. What is your data moat: do you retain rights to standardized client geodata to train models that compound, or is each engagement siloed?

What they build

BiaTech builds an application-layer AI stack for energy and infrastructure operations under the banner Human Anthropomorphic Intelligence (HApI), meaning AI meant to augment human decisions in high-stakes, high-complexity settings rather than automate them away. Three products: BiaLogic, digital twins of energy assets; BiaEdge, an edge-AI hardware kiosk for on-site inference; and AI Field Inspector, computer vision for infrastructure inspection. Current work streams include an AI solution for geothermal pipeline flow measurement (DOE-grant-supported) and computer-vision utility solutions for fire detection and vegetation management, with next-generation transportation-safety products slated to be added in early 2026. This is a broad, horizontal Physical-AI and digital-twin platform, which is both its ambition and its weakness: it spreads a small team across digital twins, edge hardware, and multiple CV verticals rather than dominating one wedge. Note this is the only company in the cohort that is energy/infrastructure-focused rather than mineral-exploration-focused; its inclusion is via the adjacent industrial-AI lens, not core mining.

Team

The two load-bearing pedigree claims both check out with framing caveats, unchanged from prior verification and reconfirmed here. Founder and CEO Nathaniel Hartwig was VP of North America for SLB (Schlumberger) New Energy in 2022-2023, a real but brief roughly one-year VP tenure, preceded by ExxonMobil (1999-2008) and McKinsey (2008-2013), a genuinely strong operator resume. A second team member, Rick Gillis, is also ex-SLB / MI-SWACO. The ex-Equinor-VP claim resolves to Sarah Delille, a genuine senior Equinor VP of roughly 16 years across multiple VP roles, but she sits as an independent board director rather than an operating executive, and is current rather than former Equinor. So both headline claims are true but dressed up: the SLB VP tenure was short, and the Equinor VP is a board member billed as leadership. Total headcount roughly 13. No public evidence of a deep in-house engineering bench comparable to MinersAI's; the profile skews senior-industry and mission-messaging over disclosed builder depth.

Funding

Capital is small and substantially non-dilutive rather than venture-validated. Reported totals conflict wildly across trackers, from about 199.6K to roughly 2.0-2.2M, and the only clearly tracked event is a DOE/OSTI grant plus prize money dated February 2024, supplemented by incubator support from Embarc Collective and Greentown Labs. BiaTech also joined NVIDIA Inception (a startup-support program, not funding). The realistic read is low single-digit millions at most, with much of it grant and prize capital rather than a priced equity seed, and well under any cap. There is no evidence of a named institutional venture lead or a disclosed priced round. This is the softest cap-table story of the four: grant-scaffolded, not market-priced.

Traction

Traction remains entirely self-reported and unverified by third parties. The company cites 500-plus terabytes processed, 7-plus use cases, and 15 percent productivity improvements, but names no marquee paying customer and discloses no revenue. As of the latest signals it is still seeking operators open to pilot the technology, which after founding in 2023 is a telling sign that design-partner conversion is slow. The DOE geothermal-pipeline-flow-measurement work is grant-scoped rather than a commercial contract. So while the product surface is broad and the pedigree senior, there is no externally verifiable customer, deployment, or revenue to anchor the traction claims. Compared with VOLUNA's live field pilot or MinersAI's named junior pilot, BiaTech's evidence base is the thinnest of the four.

Market and competition

BiaTech is the outlier that barely overlaps the other three. MinersAI, Mineflow and VOLUNA all attack mineral exploration; BiaTech attacks energy-asset digital twins, edge AI, and infrastructure inspection, competing instead with industrial-AI and digital-twin incumbents (the AVEVA/GE Digital/Cognite/Palantir-Foundry-adjacent world plus a long tail of CV-inspection startups). That is a large but brutally competitive and incumbent-heavy market where a 13-person grant-funded startup with no named customers is a very small fish. Within this cohort it wins on nothing measurable: it has less differentiated technology than VOLUNA, a thinner data moat than MinersAI, and less focus than Mineflow. Its only comparative asset is founder-CEO industry access from the SLB/Exxon/McKinsey pedigree, which could open energy-major doors, but that access has not yet produced a disclosed commercial win.

Bull case

• Genuinely senior, verifiable industry pedigree (ex-SLB New Energy VP, ex-ExxonMobil, ex-McKinsey) that can open doors at energy majors.
• Non-dilutive DOE grant funding plus NVIDIA Inception and reputable incubators (Embarc, Greentown) de-risk early burn without diluting founders.
• Broad Physical-AI/digital-twin product surface addresses a large, real industrial market with multiple monetization wedges.
• An Equinor VP on the board (even if over-framed) is a credible industry relationship and potential channel.

Bear case

• Traction is 100 percent self-reported with no named paying customer, no revenue, and the company still openly recruiting pilot operators two-plus years in.
• Grant-scaffolded cap table, not venture-priced; no institutional lead and conflicting funding data suggest a very early, unvalidated raise.
• Product focus is spread thin across digital twins, edge hardware, and several CV verticals with a 13-person team.
• Competes in an incumbent-dominated digital-twin/industrial-AI market with no visible defensible moat.

Key risks

• Design-partner-to-paying-customer conversion may never happen; the pilot-seeking posture after two years is a warning sign.
• Grant dependence exposes the company to funding cliffs if DOE and prize money are not replaced by revenue or a priced round.
• Horizontal product sprawl could prevent the team from winning any single vertical.
• Pedigree-led selling may not overcome the lack of reference customers in a proof-heavy industrial market.
• Generic name and identity collisions complicate market presence and diligence.

Questions for the founder

1. Do you have any paying customers today, and what is current revenue versus grant income?
2. What priced equity round, if any, has closed, and who led it, separate from DOE grants and prizes?
3. Which single product (BiaLogic, BiaEdge, or AI Field Inspector) has the most traction, and why not concentrate there?
4. How do the self-reported metrics (500-plus TB, 15 percent productivity) map to specific named deployments?
5. What is Sarah Delille's actual involvement and does the Equinor relationship convert to a commercial channel?

What they build

Mineflow builds AI models that predict the shape and location of mineral deposits from a site's own exploration data, aiming to cut the number of drill holes needed. A geologist uploads any dataset from an exploration site and Mineflow trains a custom deep-learning model that outputs 2D overhead and 3D volumetric predictions of deposit geometry. The technical thesis is multimodal data fusion plus cross-site learning: models get better by learning across many sites, edging toward foundation-model-for-the-subsurface territory long term. It has also shipped a free companion product, Mineflow Drive, for fast exploration-data management, which auto-infers project, location and target resource from uploaded files and, per the company, cut median processing time about 81 percent, from roughly 21 minutes to about 4. Mineflow Drive is a smart top-of-funnel wedge: give away the data-wrangling layer to seed adoption, then upsell the paid predictive models. The core bet is narrow and technically legible: be dramatically more accurate at deposit-shape prediction than anyone else.

Team

Effectively a two-person company built around a strong solo technical founder. Ryan Goggins holds a BS in Artificial Intelligence from Carnegie Mellon's School of Computer Science, where he also helped teach graduate-level Deep Learning and Search Engines courses, and spent two years at Google building ML for Display Ads optimization, claiming products and models that generated roughly 125M in ARR impact. That is a genuine, high-signal ML pedigree. The gap remains the same and is significant: there is no geology or mining domain expert on the founding team, which is a real liability selling deposit-geometry predictions into a skeptical, geology-led industry. The company is hiring aggressively for generalist software engineers and a founding ML engineer (with a below-market salary band that signals tight capital), so the team is still forming. Execution and key-person risk are concentrated in one person. The bench is the thinnest of the four; the founder quality is among the highest.

Funding

Confirmed at roughly 500-590K, raised across about seven investors: Y Combinator, Decacorn Capital, Karman Ventures, Leonis Capital and others, from the YC S24 batch. No valuation disclosed and no priced Series A found. This is the smallest and earliest cap table in the cohort, consistent with a two-person YC-stage company, and definitively under any cap. The sub-market founding-engineer compensation band corroborates that capital is tight and that the company is running lean between YC and a first institutional round. There is no evidence of a strategic corporate backer analogous to MinersAI's BHP Invent, which leaves Mineflow more exposed on both validation and follow-on capital.

Traction

Traction has improved beyond the prior dossier's single lithium trial. The headline claim remains that in early trials with a lithium miner Mineflow predicted hard-rock lithium deposit shape more than an order of magnitude more accurately than competitors. The newer and more concrete data point: Mineflow says it has delivered predictive models spanning exploration juniors up to billion-dollar mines, and cites one North American operation generating about 400M annually whose system saw grade-control error cut by 85 percent. That is a materially stronger, if still unnamed and self-reported, proof point than a single early trial, and it suggests movement from pure exploration into the higher-value grade-control/production use case. Mineflow Drive adoption is described as strong. Still, no customer is named, no revenue is disclosed, and the accuracy claims are vendor-reported without independent benchmark, so this remains promising-but-unverified traction.

Market and competition

Mineflow is the closest direct competitor to MinersAI in this cohort, both are software-on-existing-data plays, but where MinersAI is broad (standardize everything, plus marketplace) Mineflow is deliberately narrow (be the best at deposit-shape prediction). That focus is its edge and its risk. Against the funded field it is a minnow: Earth AI (~20M Series B) drills its own AI-generated targets, GeologicAI (~44M Series B) pairs core-scanning hardware with AI, and KoBold (~537M, ~3B valuation) fuses data at scale and develops mines. All of them out-resource Mineflow by one to three orders of magnitude and several own the discovery value chain Mineflow only informs. Mineflow's defensible thesis is cross-site learning: if its models genuinely compound across every site they touch toward a subsurface foundation model, a small team could build a real data moat that the asset-heavy players, siloed inside their own projects, cannot. The grade-control expansion also moves it toward stickier production revenue and away from lumpy exploration budgets.

Bull case

• Elite, high-signal solo founder (CMU AI, ex-Google-Ads ML with claimed ~125M ARR impact) attacking a legible technical problem with a clear accuracy metric.
• Genuinely differentiated claim (order-of-magnitude better deposit-shape accuracy) plus a concrete new proof point (85 percent grade-control-error cut at a ~400M/yr North American mine).
• Cross-site-learning thesis could compound into a real subsurface data moat that asset-heavy, siloed rivals cannot replicate.
• Mineflow Drive is a clever free top-of-funnel wedge, and the grade-control use case moves the company toward stickier production revenue.

Bear case

• Effectively a one-person company with no geology or mining domain expert on the founding team, selling into a conservative, geology-led industry.
• Smallest cap table in the cohort (~590K) with no strategic backer and a below-market hiring band, signaling capital and follow-on risk.
• All accuracy and traction claims are self-reported and the flagship customers are unnamed; no independent benchmark exists.
• Vastly out-resourced by KoBold, GeologicAI and Earth AI, several of which capture the discovery value Mineflow only informs.

Key risks

• Key-person and execution risk concentrated entirely in the solo founder.
• Lack of in-house geology credibility could stall enterprise adoption in a proof-heavy, relationship-driven industry.
• Order-of-magnitude accuracy claims may not survive independent, blinded benchmarking.
• Capital tightness could force a down round or stall hiring before the data moat compounds.
• Larger, better-funded players could replicate the narrow deposit-shape capability as a feature.

Questions for the founder

1. Who are the named lithium-trial and the ~400M/yr North American grade-control customers, and were those paid engagements?
2. Has the order-of-magnitude accuracy claim been validated by any independent or blinded benchmark?
3. How real is cross-site learning: do models actually compound across customers, and do you retain the data rights to make that a moat?
4. How are you overcoming the absence of geology domain expertise, through hires, advisors, or customer geologists in the loop?
5. What is current runway, and what milestones gate your next raise given the below-market hiring band?