The dossier production line, not a consultancy
Lawsone is a platform that manufactures dossiers the way a plant manufactures product: regulated input in, validated output out. An in-house agent pipeline reads public regulatory records, reconstructs candidate formulations, runs them through simulation gates, and writes the file a licensee takes to its first batch. Every dossier on the line is produced by the same pipeline, graded by the same provenance standard, and cleared by the same simulation gates.
How the line runs / record to licensable file
Agents collect the public regulatory footprint of a marketed product: label compositions, inactive ingredient limits, regional disclosure documents, interview forms, dissolution methods, and stability language. Every extracted fact lands in the record with a citation.
A second set of agents resolves the candidate formula: excipient identities and grades, quantitative ranges consistent with regulatory caps, and a manufacturing procedure on conventional equipment. Unlabelled parameters are estimated as a latent composition model, never invented as false precision.
The candidate must clear the gates before it becomes a dossier: dissolution similarity against the reference profile in compendial media (f2 >= 50), virtual bioequivalence screening, PBPK exposure estimation where the compound supports it, and a Monte-Carlo robustness sweep on critical parameters.
The final agent pass writes the deliverable: master formula, manufacturing procedure with CPPs and in-process controls, specifications, the equivalence evidence package, and the provenance map that grades every fact verified, inferred, or hypothesized.
The engine / agent stack on frontier models
The pipeline runs on an agent stack the company develops in-house. Frontier language models drive the agents that extract regulatory facts, reason about formulation constraints, and draft dossier text; deterministic simulators score the chemistry. The stack is product, not staff: the same machinery that produced the current asset registry produces the next commissioned dossier on request.
Simulation is where the platform earns its name. Dissolution mechanics compare candidate and reference profiles; virtual bioequivalence estimates whether in vivo exposure lands in the accepted window; PBPK modeling extends that estimate across populations and dosing conditions; Monte-Carlo sweeps mark which parameters are fragile. A dossier only ships after clearing its gates.
| Gate | Question it answers | Output in dossier |
|---|---|---|
| Dissolution similarity | Does the candidate release like the reference in every compendial medium? | f2 table, overlay plots |
| Virtual bioequivalence | Does predicted exposure sit inside the BE window? | Cmax / AUC estimates, CI |
| PBPK | Does exposure hold across populations and dosing conditions? | Model file + sensitivity |
| Robustness sweep | Which parameters can drift without breaking equivalence? | Monte-Carlo report |
Specimen / what a finished dossier looks like
Compound identities stay gated under NDA, but the shape of the product is public. A finished dossier opens with the claims table below - every fact tagged so diligence knows exactly what is established and what is inference - then runs to the full formula, procedure, specifications, and simulation appendix.
| Claim | Provenance | Value |
|---|---|---|
| Reference-product dissolution profile, all compendial media | Verified | [redacted] |
| Excipient grades and supplier class | Verified | [redacted] |
| Target binder and disintegrant levels, % w/w | Inferred | [redacted] |
| Critical process parameter windows | Hypothesized | [redacted] |
| Optimal Trial-1 batch size for pilot-BE | Hypothesized | [redacted] |
Ask for the index / or commission a dossier
The non-confidential dossier index lists every asset by class and status. To commission, name the molecule, dosage form, and target markets - feasibility, timeline, and price come back first. Write to aws@lawsone.si.