Technical Lead
Amr Idlibi
Built the calibration model architecture and exploration intelligence dashboard. Robotics and autonomous systems background; runs the technical roadmap.
LinkedInFerra Labs builds the AI exploration platform that turns geologic hydrogen targeting from geological guesswork into confidence-scored investment decisions.
The problem
Clean energy is locked to geography.
Two billion people have no reliable access to clean energy. Heavy industry, including steel, cement, fertilizer, shipping, and aviation, cannot decarbonize. Global energy demand will double by 2050.
This is not because clean energy doesn't exist. Every form of it we have is locked to a specific place on Earth. Solar needs sun. Wind needs wind. Hydropower needs rivers. Even green hydrogen, the cleanest fuel we currently make, needs cheap renewable electricity, which most of the world does not have.
For the entire history of energy, we have assumed clean energy must be extracted from somewhere specific and shipped to where it is needed. That assumption is the bottleneck.
What is stimulated geologic hydrogen
The Earth has been making clean hydrogen for billions of years. When water meets iron-rich rock deep underground, it produces hydrogen gas through a reaction called serpentinization. Stimulated geologic hydrogen is the engineered acceleration of that reaction: inject water into the right rock, and the Earth manufactures fuel on demand.
Cross-section · serpentinization reaction schematic
↓ The loop
The loop
0
flow rate gap to commercial viability
Closed by stacked technical levers — chemistry, fracture density, microbial suppression, and targeting. Targeting is the lever that unlocks the rest.
The industry can't break the loop.
The science is no longer the question. Serpentinization works. Iron-rich ultramafic rock exists across roughly a third of Earth's continental crust. Operators including GeoKiln, Eden GeoPower, Vema Hydrogen, and HyTerra are building stimulation technology with serious capital backing.
The bottleneck is structural. Operators need capital to drill pilots. Investors won't fund pilots without proof of viable targets. Proof requires drilling. Drilling requires funding. The loop is unbreakable from inside.
The lever that breaks it is targeting. Operators today are drilling against public geological data gridded at 1 kilometer when the iron-rich pods that actually react are 100 meters wide. Without higher-resolution targeting, no drill plan is credible enough to unlock the capital that funds it.
The insight
Two questions. Two output heads. One system.
Every previous attempt at AI for hydrogen exploration conflates two fundamentally different questions. The data available to answer them is different. The uncertainty profile of each is different. We keep them separate.
Head 1
Asks where suitable rock exists. This is geology targeting — is there sufficient Fe2+ bearing ultramafic rock at economically viable depth at this location? It's answerable now from public satellite spectrometry, aeromagnetics, and structural data. High confidence is achievable today.
Head 2
Asks whether stimulation will produce viable hydrogen. This is yield prediction. It requires drill results and live pilot data to calibrate. Confidence intervals are wide until late 2026 and narrow with each new commercial pilot. We tell operators and investors honestly which question we're answering and what the confidence is. That separation is what makes the system credible.
If I could put my hand on a Bible and say I definitely know I have a good drill target in this spot on Earth, I wouldn't need to pre-drill. It would speed the process by years and give me greater confidence in getting funding.
Robert Dombrowski
Director of Subsurface · GeoKiln
The platform
Inputs. Model. Output.
Layer 1
Data Ingestion
Multi-resolution, native-resolution preserved
ASTER hyperspectral
USGS aeromagnetics
OneGeology bedrock
Soil gas grids
Drone hyperspectral
Layer 2
Geological Encoding
Domain knowledge before the AI sees anything
Serpentinization potential
Fe2+ availability
Fault density
Microbiome risk
Thermal feasibility
Layer 3
Physics Model
Three encoders fused, Bayesian uncertainty
CNN raster encoder
GNN fault network
Point encoder soil gas
Bayesian fusion
Layer 4
Target Package
Structured decision document per site
Suitability score
Depth confidence
Yield range
Risk flags
Pre-drill protocol
Layer 5
Calibration
Every drill result improves the model
Operator data sharing
Model retraining
Network compounding
01 · Inputs
Multi-resolution data fusion
Public hyperspectral imagery (ASTER, EMIT) discriminates surface iron oxidation state. Aeromagnetics infer subsurface magnetite as a serpentinization proxy. USGS drill core archives, OneGeology bedrock maps, and ARPA-E program data ground the model. Proprietary partner data — soil gas grids, induced polarization surveys, drone hyperspectral — sharpens it.
02 · Intelligence
Physics-informed hybrid model
A Bayesian neural network with three parallel encoders fuses gridded raster, fault network graphs, and sparse point measurements. Hard physics constraints in the loss function prevent geologically impossible predictions. Outputs are probability distributions, not point estimates. Every prediction carries explicit confidence bounds.
03 · Output
Drill target intelligence package
Per candidate site: rock suitability score with confidence intervals, depth confidence curve, yield potential range with explicit bounds, microbiome consumption risk, estimated drilling cost, pre-drill validation protocol, 45V tax credit eligibility flag, and data quality tier. This is what an operator hands to a venture capitalist.
Every drill result anywhere in the world makes the model smarter. Operators who partner with Ferra Labs early help calibrate the system that the rest of the industry will eventually run on.
The deliverable
What an operator gets.
Each drill target package is a structured decision document. It's not a heatmap. It's the document an operator hands to a venture capitalist to unlock the next funding round.
Drill target package · Target 03 · Stillwater Complex · MT
v3.6
Rock suitability
0.87 ± 0.04
Depth confidence
1,800 – 2,400 m
Predicted yield
350 – 650 kg H₂/day
Yield confidence
Wide. Narrows post-Q4 2026 pilot.
Microbiome risk
Low — formation temp >122°C
Est. drilling cost
$3.6M – $5.4M
45V credit eligibility
Eligible — Tier 1 ($3/kg)
Data quality tier
High — public + 1 partner stream
Pre-drill validation recommended
- Soil gas grid · 250m spacing · 6 weeks · ~$28K
- Induced polarization · phase 2 · 4 weeks · ~$17K
- Total: $45K · 10 weeks
Every package carries explicit uncertainty bounds. Honest uncertainty is the credibility.
Validation
Validated by the people building this industry.
Validation sources
Operators · Direct conversations
GeoKiln, Eden GeoPower, and others building stimulation technology today
Academic · ARPA-E grantees
MIT (Iwnetim Abate), Penn State (Liu/Elsworth), Colorado School of Mines
Industry · Compiled outreach
Researchers, journalists, geophysical advisors
AI for site identification is the most useful application of AI in this industry.
Mark Hansford
Subsurface Integration Lead · Eden GeoPower
AI ranking drill targets by rock volume, iron purity, and depth would give more confidence in where you're choosing to drill.
Industry Expert
Eden GeoPower
USGS Mineral Resources Program · AMAX Drill Core Archive 1969–1977 · ARPA-E Geologic Hydrogen Program · Templeton et al. 2024
Team
Team
Built by the team learning at the frontier.
TKS Innovate students working on systems that compound.
Project Manager and Operations
Uygar Ceylan
Drives product decisions and hardware integration. PCB and electronics background; bridges what the model outputs with what operators actually need.
LinkedIn
Industry and Communications
Ivana
Leads operator outreach and presentation. Owns the relationships with industry experts at Eden GeoPower, GeoKiln, and ARPA-E grantee labs.
LinkedIn
Research
Adham
Owns the research foundation. Stress-tests the geology, economics, and competitive landscape. Every technical claim on this site has passed through his review.
LinkedIn
Market and Economics
Tiara Ansari
Drives the economic and industry analysis, ensuring Ferra Labs' positioning survives investor and operator scrutiny. Based in Iran.
LinkedInThe vision
Energy manufactured anywhere from rock and water.
We exist to dissolve the geographic constraint on clean energy by unlocking the stimulated geologic hydrogen industry. By 2045, communities that have never had reliable clean energy produce their own from local geology. Heavy industries decarbonize because the molecular fuel they need is finally affordable. The geopolitics of energy is replaced by the geology of energy, which is distributed across every continent on Earth.
We are the foundational data layer that makes the industry investable.
Partnership
Be one of the first operators in the foundational data layer.
The first cohort of operator partnerships defines how the platform develops. Operators in the first cohort get early access to drill target packages, embedded technical partnership during drilling, and a seat in shaping the system that the rest of the industry will eventually run on.
We're also open to conversations with academic geochemistry programs, ARPA-E grantee labs, industry researchers, and investors thinking about the foundational data layer of stimulated hydrogen at the five-year horizon.