Labs — Scientific + regulatory intelligence for pharma, biotech & the capital behind them
Cortellis and Citeline cover the world. They lag 2-4 weeks on Indian regulatory. We fill that gap — CDSCO filings, DCGI letters, Indian Ph-3 trials, patent cliffs — plus contested-science DD on the edge cases your internal team doesn't have the time or specialist depth to run. Three case studies shipped. Every finding citation-grounded. Built to complement your Cortellis seat, not replace it.
Case 01 · live commercial
Indian oncology biosimilars. 80 PubMed papers · 66 Ph-3 trials · 22 biotech VCs indexed · 4-layer citation grounding.
Case 02 · methodology proof
LENR / cold fusion. 9,795 papers analyzed. 37-year open contested-science case. Same pipeline, different domain. Full PDF below.
Case 03 · methodology proof
LK-99 superconductor. 48 papers classified. 6-week resolution. Cu₂S artifact identified. Closed-case counterpart to LENR.
Live this week — verified Apr 24 2026
48,083 FDA Orange Book products · 16,724 active US patents · 1,980 drug-product exclusivities indexed and joined to expiry dates — refreshed monthly. 10 CDSCO India circulars ingested live for the first design-partner tenant. 249 / 249 tests green in 4.7s. Three production agents on cron — daily 09:30 IST for India regulatory, monthly for the patent cliff.
The research infrastructure
66,777
Records under monitoring
48,083 FDA Orange Book products · 16,724 active US patents · 1,980 exclusivities · 10 live CDSCO filings — refreshed monthly, dedup-safe, tenant-scoped
3
Case studies shipped
LENR (contested physics, 37yr open) · LK-99 (closed case, 6-week resolution) · Indian biosimilars (live biotech pipeline) — 4th case (Alzheimer's amyloid) queued
249
Tests passing in 4.7s
Real-DB integration tests · async conftest with TRUNCATE cleanup · scraper + service + route + agent layers — the pipeline is provable, not vibes
4-layer
Citation grounding
strict prompt + indexed source corpus [S#] + citation validator + honest insufficient-data flags — 0 broken citations in shipped reports
3-tier
LLM routing
Gemma 4 local ($0) for classification, Claude for synthesis, NVIDIA Qwen3-480B ($0) as production fallback — no OpenAI dependency
~$5
Compute per report
Versus ₹12-15 Cr/year for Cortellis + Citeline + Clarivate India seats. We complement, we don't replace — we fill their India gap.
We complement Cortellis, Citeline, and Clarivate — we don't replace them. Our wedge is India-specific regulatory granularity and specialist contested-claim depth on the questions your internal team doesn't have the bandwidth to run. Corpora per domain persist across engagements — a second run in the same therapeutic area starts from the indexed source base, not from zero.
01 / What comes back
Every engagement is fixed-scope, fixed-fee. A written deliverable in your hands in two weeks, not six. Defensible under IC scrutiny.
01 / Output
Two weeks. One asset. One verdict.
Past the pitch deck. Published science evaluated at PhD depth with four-panel adversarial review. Competitor pipelines mapped. Artifact risks flagged with named mechanisms, not vague skepticism. One written verdict with cited sources — every claim traceable to a source corpus.
Deliverable
Diagnostic report + plausibility score + IP + regulatory path + team + market — 10-17 pages, citation-grounded.
02 / Output
India-specific where the global tools lag 2-4 weeks.
Cortellis and Citeline miss Indian regulatory granularity — CDSCO filings, DCGI approval letters, biosimilar import licences. We scrape, classify, and monitor four CDSCO surfaces (notifications, approved drugs, biologicals, clinical-trial permissions). Live ingestion is running for design-partner tenants — first April 2026 CT-05 / CT-10 / CT-12 circulars are already in the database.
Deliverable
Daily alerts + monthly digest + on-demand competitive summary — JSON, markdown, or PDF, your choice.
03 / Output
FDA Orange Book × your portfolio. 24-month window, monthly refresh.
48,083 FDA Orange Book products and 16,724 active US patents indexed and joined to the cliff date. Filter by therapeutic area or originator (Otsuka 45 patents cliffing, Takeda 37, Bausch 24). Substring-match against your CDSCO drug-name register — surface every patent expiring in your filing window. Built for Indian generic formulators timing ANDA submissions for Day-1 entry.
Deliverable
Monthly markdown / PDF report + JSON feed + tenant-scoped match list against your filings.
04 / Output
Pick a therapeutic area, a contested claim, or a pipeline thesis.
If the domain isn't one we've shipped publicly, we scope a custom engagement. Biotech, pharma, materials, contested physics — the pipeline generalises. Ten days to a written deliverable, NVIDIA-tier compute cost, citation-grounded output with an honest 'insufficient data' flag where the corpus is thin.
Deliverable
10-day turn · 10-15 page report + source corpus + dashboard (for repeat engagements).
Case study 01 / Live commercial · Indian pharma
A pitch-ready oncology gap analysis built on the same pipeline adapted for biotech. PubMed + ClinicalTrials.gov + Indian pharma entity universe + biotech VC index. 4-layer citation grounding — every factual claim traceable to an indexed source. This is the template for the pharma CI wedge: India-specific, cite-defensible, 10-day turnaround.
80
PubMed papers
66
Indian Ph-3 trials
22
Biotech VCs indexed
20
Pharma companies
11
In-text citations · 0 broken
33 s
NVIDIA Qwen gen time
4-layer citation grounding
Grounding prompt + indexed source corpus with [S#] references + citation validator + honest insufficient_dataflags where the corpus is thin. Caught an LLM hallucination ("10,000+ China oncology patents" when the DB had zero) and re-engineered. Reports ship with 0 broken citations.
The CDSCO wedge
Cortellis and Citeline have global coverage but lag Indian regulatory granularity by 2-4 weeks. CDSCO filings, DCGI approval letters, state-FDA inspection reports. That gap is the moat — foreign tools don't do it, generic Indian tools don't have the AI pipeline. We build into that gap.
Live patent-cliff feed
48,083 FDA Orange Book products + 16,724 active US patents indexed and joined to expiry dates. Per-tenant cliff calendar matched to your CDSCO drug-name register. View live dashboard →
Two methodology proofs follow — same pipeline, different domain
LENR (open contested-physics, 37 years unresolved) and LK-99 (closed in 6 weeks, Cu₂S artifact identified). Demonstrating the source→classify→synthesize→deliver stack on harder corpora than biotech. Not the commercial play — the craft proof.
Case study 02 / Methodology proof · contested physics, open
We analyzed the entire published literature on Low Energy Nuclear Reactions — 37 years of contested science that mainstream physics dismisses and believers oversell. Four-panel adversarial methodology. Every finding stress-tested from both sides.
9,795
Papers analyzed
396
Experiments graded A-F
1,774
Patents mapped
223
Researcher profiles
193
Cathode recipes clustered
10,300+
Total across 4 domains
~$5
Total analytical cost
Signal co-occurrence paradox
Transmutation co-occurs with excess heat in 67 papers vs 11 for helium-4. The field's theoretical mainstream models the minority pathway. Never formally published before this analysis.
Chemistry ruled out
PdH formation enthalpy verified at 37 kJ/mol (Flanagan & Oates 1991). Total chemical energy for a typical cell: ~2 kJ. Reported excess heat: 1,300+ kJ. Chemistry is orders of magnitude too small.
Adversarial methodology
Four-panel review: steelman, attack, blind-spot, and proponent-calibrated agent. Probability estimate: 10-30% genuine nuclear phenomenon. Three independent methods converge.
Cathode recipe analysis
193 cathode recipes — first systematic ML analysis ever performed on LENR preparation data. Powder/nano cathodes succeed at 3x the rate of thin films. Heat treatment is the only prep that improves outcomes (+14%).
COP by calorimetry method
Better instruments produce lower results. Flow calorimetry: median COP 1.24. Isoperibolic: 1.33. Pd + flow (most rigorous): 1.20. The 'real' effect, if it exists, is at COP ~1.2.
Contested science pattern match
Scored LENR against 17 historical cases — continental drift, H. pylori, prions, polywater, N-rays. LENR scores 6/20 on vindication factors. Closer to polywater than any vindicated case.
Closed-case counterpart below
LK-99 — same pipeline applied to a resolved-in-six-weeks claim. Different regime, same source→classify→synthesize→deliver stack. The pair (open + closed) demonstrates the full range.
Case study 03 / Methodology proof · superconductivity, closed
Same pipeline, different regime. The July 2023 Korean room-temperature superconductor claim went from arXiv posting to community rejection in 6 weeks. We built the post-mortem: 48 papers classified, artifact named, resolution mechanism mapped, Langmuir pathological-science score applied.
48
Papers classified
6 wks
Claim → consensus
Cu₂S
Artifact identified
4 / 6
Langmuir pathological
4-factor
Resolution-speed framework
10 pp
Full report PDF
The artifact, named
Cu₂S impurity at ~1.5% of sample mass, with a first-order phase transition at ~377 K. That transition produces both the "resistance drop" and the "half-levitation" signal the original team interpreted as superconductivity. Textbook mechanism, well-documented — once you looked.
Verdict distribution
4 supports, 25 neutral, 12 contradicts, 7 identifies-artifact. All 4 supports came from the original team or earliest replicators. By end August 2023, supporter base in condensed-matter physics was effectively zero.
4-factor resolution speed
Synthesis cost (low), artifact specificity (Cu₂S, textbook), institutional engagement (MPI, LBNL, Beijing active), venue openness (arXiv). All four low-friction → 6-week close. Same framework applied to LENR: all four high-friction → 37 years open.
02 / Who we serve
CI teams at top-10 Indian pharma, biotech VCs, corporate R&D, family offices, and contested-science evaluators. Anyone whose next decision depends on Indian regulatory granularity or specialist scientific depth.
Pharma competitive intelligence teams
CDSCO + DCGI filings 48+ hrs ahead of manual pull · Indian Ph-3 competitor monitoring · patent cliff calendar · biosimilar landscape
Biotech VCs (India-focused)
Scientific DD on portfolio candidates · gap analysis on emerging therapeutic areas · Indian regulatory trajectory · 22 peer-VC intel indexed
Corporate R&D + business development
Competitive pipeline monitoring · in-licensing candidate evaluation · prior-art defensibility · Indian clinical site intel
Family offices + allocators
Therapeutic-area thesis builds · co-investment screening on biotech/pharma deals · management-grade contested-claim risk memos
Contested-science evaluators (labs, reinsurance, foundations)
Independent artifact-risk analysis · resolution-speed scoring · specialist literature reviews (LENR, superconductivity, novel claims)
03 / Methodology
Bulk extraction at $0
Gemma 4 running locally processes thousands of papers in batches — 487 mainstream Pd-H papers in 6 hours, 9,795 LENR papers in 42 hours. Total cost: zero. We read 2,000 papers in a week for the cost of one junior analyst month.
Cross-domain verification
Every finding is cross-referenced against mainstream physics, meta-science failure modes, and 17 historical cases of contested science. We caught a 3-4x enthalpy error that would have shipped in any single-source analysis.
Four-panel adversarial review
Every major finding goes through a steelman agent, an attack agent, a blind-spot agent, and a proponent-calibrated agent. No claim ships at a confidence level it cannot support. The LENR probability estimate (10-30%) came from four independent methods converging.
Structured outputs, not just a PDF
Every engagement produces a graded dataset alongside the written report — experiments, patents, COP measurements, researcher network. Structured so a second run in the same domain compounds rather than starts fresh. Delivered as PDF + CSV + JSON; a queryable dashboard is available on request for repeat engagements.
Every engagement produces a citable, defensible artifact. Every source is tracked. Every finding is reversible.
Engage
Pick a therapeutic area, a contested claim, or an India-specific regulatory question your internal team doesn't have the specialist depth to close. Ten days, fixed scope, citation-grounded deliverable, NVIDIA-tier compute cost. If the output doesn't pay back an hour of your Associate Director's time, we refund. Sample our work below.
Free · monthly · lead-gen
CDSCO + Indian biotech digest. Five bullets, first Monday of each month. India-specific filings, Ph-3 moves, competitive intel your Cortellis seat missed. No pitch, no pricing page, just the intel.
Paid · monthly · live this week
Patent cliff calendar feed. 48K FDA Orange Book products, 16,724 active patents joined to your CDSCO drug-name register. 24-month forward window. Built for Indian generic formulators timing ANDA filings for Day-1 entry.
Three cases shipped · LENR · LK-99 · Biotech · plus the live Patent Cliff Calendar feed · domain pickable on commissioned DDs