MedSynth

Spatial, agentic drug discovery — designed in augmented reality, scored by AI, sharpened by federated clinical data.

Start the pitch ↓← Back to Larynx

The pitch

MedSynth™, in nine slides.

01 / 09

The problem

Drug discovery still runs on flat, siloed tools.

01 · Flat

2D for a 3D problem

Binding happens in three dimensions. Most design still happens on a flat screen, one molecule at a time.

02 · Siloed

Disconnected engines

Generation, docking, and free-energy tools live in separate silos that rarely share a scoring view.

03 · Blind

Cut off from outcomes

Scoring functions can't safely learn from the clinical and real-world data locked inside hospitals.

Adam Gabriel, Polyminventor™01 / 09

What MedSynth™ is

One platform from target to candidate — in space.

MedSynth™ builds a 3D representation of a biological target, proposes and re-scores candidate molecules with agentic AI, and lets scientists review the whole design session in augmented reality.

Spatial

Designed in 3D

Review targets and candidates in AR on visionOS, with a desktop fallback.

Agentic

Scored by agents

A workflow orchestrates potency, toxicity, and synthesis predictors.

Federated

Tuned by reality

Re-weighted by real-world outcomes without moving patient data.

Adam Gabriel, Polyminventor™02 / 09

How it works

Four stages, one continuous loop.

Represent the target

Build spatial, electrostatic, pharmacophoric, and accessibility features of the binding site.

Propose candidates

A generative or search model proposes therapeutics, conditioned on the indication.

Score, multi-objective

Rank on predicted potency, toxicity and off-target risk, and synthesizability.

Review & modify

Scientists edit candidates in AR or 2D; edits are re-scored and fed back into the loop.

Adam Gabriel, Polyminventor™03 / 09

Agentic scoring

An agent runs the bench work.

A workflow manager orchestrates the calls a medicinal chemist would make by hand — and searches the design space far faster.

Potency, toxicity, and synthesis-difficulty predictors
Specialized scores: β-lactamase resistance, kinase off-target, cardiotoxicity, IP novelty
Monte Carlo tree search and RL policies over structural edits

+Binding / potency — predicted affinity for the target pocket.

−Toxicity / off-target — penalized before synthesis.

≈Synthesizability — can the lab actually make it.

Adam Gabriel, Polyminventor™05 / 09

Federated learning

Learn from the clinic. Move no records.

Remote data custodians compute model updates on clinical, omics, and real-world data that never leave their walls. MedSynth™ aggregates the updates — never the patient records.

Stays local

Data never moves

Hospitals keep patient-level records; only aggregated model updates are shared.

Re-weighted

Scored by outcomes

Toxicity, efficacy, resistance, and stratification scores tuned to real-world signals.

Orchestrated

Built to govern

A federation service distributes parameters and aggregates gradients, by design.

Adam Gabriel, Polyminventor™06 / 09

Indications

One platform, conditioned per modality.

Antibacterial

β-lactam and β-lactam-inspired scaffolds for penicillin-binding-protein pockets.

Antiviral

Small molecules against viral proteases, polymerases, and capsid proteins.

Oncology

Kinase inhibitors and protein–protein interaction disruptors.

Immunology

Modality-specific generation across immunology and metabolic disease.

Antibody / protein

Generate and re-score protein and antibody therapeutics for design teams.

+ more

Indication metadata conditions generation — new modalities slot into the same loop.

Adam Gabriel, Polyminventor™08 / 09

Who builds it

A small senior team that ships.

Larynx™ is deliberately small. Two people who have built and shipped real systems own every engagement directly — strategy, architecture, and the code itself.

Founder

Adam Gabriel

Sets the technical direction and runs every engagement end to end — from first scope to production handoff. Founded Larynx to build software the way a senior team would, without the agency overhead.

Founding engineer

Eric Lee

Owns architecture across the stack, from the model layer down to the infrastructure under it. Leads the engineering behind MedSynth™'s spatial, agentic, and federated systems.

Let's talk

Build the future of discovery with us.

info@thelarynx.com →