The licensed-data
marketplace
for scientific AI.
Scientists sell the rights-clean experimental data locked in their silos — unreleased structures, raw diffraction, negative results — to the AI labs that now need it. Brokered on the doi.bio knowledge graph.
The Problem
The most valuable science data never leaves the lab.
Released PDB is already free and CC0. The value is in what's not scrapable — and today it's trapped, unlicensable, earning nothing.
"We provide ways for scientists to unlock the value of the data in their silos." — market.doi.bio, founding idea
- Scientists sit on unreleased structures, raw diffraction, assay panels and negative results — real money, with no way to license or sell it.
- AI labs need clean, provenance-tracked, indemnified science data to train on — and can't buy it. No marketplace exists.
- Pharma already trades data in $100M–$1B+ bilateral deals — one-off, lawyer-by-lawyer. No rails, no liquidity, no long tail.
The data exists. The market doesn't.
Why Now
Licensed data just became mandatory.
$1.5B
Bartz v. Anthropic settlement, Sept 2025 — pirated training data is now a balance-sheet liability
EU AI Act
Training-data provenance now required by law — provenance is no longer optional
0
Data marketplaces built for science — a dozen exist, all news / text / media
- Clean rights are now the only safe input. Labs increasingly need provenance-tracked, indemnified data — and a marketplace that guarantees it is worth more than the raw bytes.
- Every AI-data marketplace is news & text. TollBit, ProRata, Human Native, Cloudflare — all built tollbooths for media. Science is completely unserved.
Everyone built tollbooths for news.
Nobody built the rails for science.
The Solution
A provenance-first marketplace for science data.
Biochemists and biotech/pharma sell rights-clean data to AI labs — and to each other — with clean licenses, indemnity and gated delivery built in. We hold the metadata and broker the deal; the value flows to the scientist.
→
→
→
04
Gated, watermarked download
- Rights-clean by construction — seller attestation, standard license tiers, indemnity, optional verifier badges.
- Gated & watermarked delivery — AES-256 encrypted at rest, metered downloads, a per-order canary in every file.
- API-first core — humans, autonomous agents (MCP) and a bulk uploader are all clients of one headless product.
# flagship listing
Helix — Unreleased KRAS G12D
co-crystal structures
price $48,000
rights ✓ verified · indemnified
deliver gated · watermarked
For Scientists
Your silo is an asset.
Every biochemist has a hard drive of data that took years and grant money to produce — and currently earns nothing. We turn it into licensable inventory, cleanly and safely.
Structure
Unreleased structures
Embargoed PDB entries and co-crystals — the exact non-scrapable data labs pay most for.
Raw
Diffraction & raw instrument data
Structure factors and raw images — richer than any deposited coordinate file.
Assays
Assay & screening panels
Proprietary compound-target and screening data from lab and ex-pharma vaults.
Negative
Negative & failed results
The experiments nobody publishes — and exactly what models need to learn what doesn't work.
Preprint
Preprints & unpublished notes
Pre-publication findings and lab notes, licensed ahead of the journal.
Vaults
doi.bio knowledge vaults
Curated, annotated markdown vaults — human-scored knowledge, ML-ready.
Unlock the value locked in your data.
The doi.bio Graph
A metadata catalog of all of biochemistry.
market.doi.bio sits on doi.bio — a pre-connected knowledge graph that already maps the entire field: every protein, structure, paper, researcher and company, wired together as plain markdown and summarized for machines.
8M+
Full-text articles (PMC)
104K+
Unified person records
133K+
Indexed graph entities
Not a marketplace with a search box.
A marketplace that already knows the whole field.
The Moat — The Sourcing Engine
Tell us what you need. We know who has it.
When a buyer's request can't be filled from the catalog, the graph ranks who most likely holds the data — on-platform to buy now, or off-platform to warm-recruit. No generalist marketplace can do this. They have no scientific graph.
RFQ "p53-stabilizing co-crystal,
unpublished"
→ ranked holders Schepartz · Fersht …
→ 0 shared keywords · semantic match
Every unfilled request is a
proprietary demand signal.
PDB structure → UniProt protein290,641
PDB structure → PMID paper204,213
Scholar → publications57,188
Protein → AlphaFold model53,171
One traversable graph — the moat133,626
request → source → new supply → richer catalog
Traction — Built & Live
A working product. Built solo, end-to-end.
Live at market.doi.bio — full stack (API core, web app, agent server) deployed and reboot-safe. Payments run in mock mode pending live Stripe keys and counsel; everything else is real and verified.
Live
Catalog + semantic search
Listings with license tiers and a per-order license snapshot; embedding-based discovery.
Live
Stripe Connect payments
Mock by default; real-mode webhook → fulfilment verified. Platform fee split to sellers & verifiers.
Live
Gated, encrypted, watermarked
AES-256-GCM at rest, metered download, per-order canary in every delivered file.
Live
Seller contribution flow
Create → upload → publish, Connect-gated paid listings, an Open Access free tier.
Live
Semantic sourcing + warm-invite
Graph match → prefilled invite → recruit off-platform holders into the deal.
Live
MCP agent server
Autonomous agents search, post RFQs, and purchase — the market is agent-native.
Agents don't just search the market. They buy.
Business Model
We take a cut of every clean transaction.
Free data (PubMed, PDB, PMC) is the lead-gen funnel. Paid proprietary data is the margin. Biotechs sit on both sides — one population seeds supply and demand, beating the marketplace cold-start.
Free data brings them in.
Proprietary data pays.
Marketplace fee — every sale10–20% / txn
Verification badges — quality + rightsfee / dataset
Sourcing & brokerage — the moatdeal fee
Managed storage & clean-roompremium
Recruiter & broker networkcarved from fee
Market
A market forming right now.
AI training-data market 2033 (projected)$16B+
AI training-data market 2026$3.9B
The science slice our high-value wedgethe wedge
~4× growth in seven years — and the science slice commands the highest price per byte.
Demand is already proven — pharma trades data bilaterally, with no marketplace to sit in between:
Incyte × Genesis Therapeutics>$1B
Illumina — Billion Cell Atlasflagship
Mayo Clinic × Merckmulti-yr
Today: one-off, lawyer-by-lawyer→ us
Why Us
Why this compounds — and can't be cloned.
01 Core
The scientific graph
Anyone can build a checkout. Nobody else has 133K linked entities across all biochemistry — the sourcing engine that fills requests the catalog can't.
02 Timing
Provenance is law now
Bartz + the EU AI Act make rights-clean, indemnified data mandatory — exactly what we broker.
03 Liquidity
Biotechs on both sides
One population buys and sells — the cold-start killer that generalist marketplaces never solve for science.
04 Flywheel
The demand log
Every unfilled RFQ is private signal: request → source → new supply → richer catalog. It compounds.
05 Native
Agent-first architecture
Built API-first with an MCP server. When agents become the buyers, we're already their storefront.
06 Founder
Founder-market fit
Built the graph and the marketplace solo, with CASP-validated scientific judgment. A rare pair.
Founder
The receipts.
Steven Ness — PhD, Computer Science. Solo founder. Built the doi.bio graph and the live market.doi.bio marketplace end-to-end: ingestion, semantic search, payments, gated delivery and the agent server.
Builds the whole stack —
and knows what science is real.
CASP15Ranked #1 against all 26 competing groups on target T1124 — a blind, refereed, world-level structure-prediction benchmark.
SOLOBuilt doi.bio (114K+ files, sub-second search) and the full marketplace stack — server, web app and agent server — alone.
The Ask
Raising $250K pre-seed.
Enough to turn a live product into a business with real deals and clean legal rails.
Solo founder · Steven Ness
1
Legal railsCounsel to make the license pack + indemnity binding, plus storage ToS/DPA — the gate to real paid deals.
2
First dealsRecruit design-partner buyers to post real RFQs; warm-invite the first sellers; close the first biotech → AI-lab transaction.
3
Keep buildingLive Stripe payments, S3 + KMS managed storage, and pgvector for semantic sourcing at scale.
Scientists own the data
AI needs. We built
the rails to sell it.
Provenance-first. Agent-native. Live.