Ryogena

Your compound data knows
more than your team does.

Drug discovery generates more signal than any team can manually interpret. The insight that advances your program already exists in your data. Finding it is the problem Ryogena was built to solve.

Introducing our first product pico ⁻¹²
What Pico does

The AI copilot built for the DMTA cycle
in small molecule drug discovery.

The signal

Your SAR table already contains the answer to your next synthesis decision. Pico reads across your compound library, your assay history, and your failed series to surface the pattern your team would take three more weeks to find manually. The insight was always there.

The workflow

Design. Make. Test. Analyze. Pico lives inside every stage — bolted onto your ELN, running alongside your chemistry, not instead of it. The DMTA cycle gets faster without changing how your team works.

The data

Everything stays on your infrastructure. No compound leaves your server. Pico runs locally — on-device, on-premise, fully air-gapped if your environment requires it. Built for regulated environments from day one.

The result

Fewer cycles. Better candidates. The program that was 18 months away becomes 12. The series you would have abandoned surfaces a hidden scaffold hop. The data sitting in your ELN starts working for you.

Why Ryogena
I.

Built by people who ran discovery programs

Not by engineers who read about drug discovery. By someone who sat at the bench, ran SAR campaigns, and felt the daily cost of disconnected data.

II.

Your data never leaves your hands

IP is everything in pharma. Pico runs on your infrastructure. The model is local. The data is yours. Always.

III.

AI that respects the science

Pico augments medicinal chemistry judgment — it doesn't replace it. Every prediction carries applicability domain scoring, uncertainty flags, and provenance.

IV.

Compliance from line one of code

ALCOA+ audit trails. 21 CFR Part 11 environments. SOC 2 ready at release. Regulatory rigor was baked in — not bolted on.

How we build

A small studio shipping production code at the pace of a venture-backed team — without the team.

Ryogena is a one-person studio that builds AI-native scientific software with Claude as a co-author. The output is regulated-industry production code, shipped solo, audit-ready from line one. The discipline below is what makes that possible — and is the same discipline we apply to every product Ryogena builds, not just Pico.

I · The .md system

A constitution the AI cannot ignore

Beyond the standard CLAUDE.md and MEMORY.md, every product ships with Architecture Decision Records, operational runbooks, and compliance-mapping artifacts — the documentation discipline of a twelve-person team, authored by one and the AI.

II · Tests-first, always

No code lands without a failing test in front of it

Backend logic is a pytest case before it is a function. Frontend pure logic is a Vitest spec before it is a hook. User-facing flows are a Playwright spec before they are a button. Red → green → deploy. The test suite is the audit log of intent.

III · MCP-native by design

Every product ships with a Model Context Protocol surface

If a tool exists in the product, it exists in MCP. Scientists call it from Claude Desktop; Anthropic Claude calls it from anywhere. We built one for Pico, and we open-sourced ryogena-pubmed-mcp so others can study the integration pattern.

IV · Version control as audit trail

Git history is the ALCOA+ paper trail

Every commit signed and co-authored. Branch-protected main. 24-hour cooling window before merge. Feature branches per concern, not per session. The git log answers who, what, when, why for every change — the same questions a regulator would ask.

V · Compliance is architecture

ALCOA+ and SOC 2 are decisions, not retrofits

Soft-deletes only — never DELETE. emit_audit() on every state change. SMILES canonicalized before storage. SSO/SAML at TEAM tier and above. These are not features added before an audit; they are defaults from line one of code.

VI · Hand-off review every time

Every change ships with a smoke test and an honest pushback

Every shipped feature comes with: what changed, how to use it, three numbered smoke-test steps the operator can run on prod, and an honest note on what's brittle. Half-finished claims get caught at the bench before they reach the customer.

Ryogena Council
How we build, automated

A HITL agentic panel of specialists QC every line before it ships.

Pico is built with Ryogena Council — an in-house review platform that turns every git commit into an automated board meeting. A pre-commit hook routes the staged diff to a panel of specialist agents that monitor changes in parallel, score it against their discipline, and post their findings before the line moves. Nothing reaches main without a verdict. Safe, accurate, compliant code is shipped, any flaw is tagged and re-written, any insecurity is handled by a human. Ryogena = Clarity from the darkness.

I Pre-commit dispatch. The staged diff is sliced and routed to the agents whose charters match. No noise, no blanket reviews — each commit gets exactly the readers it needs.
II Parallel panel. Every queued agent reads at the same time, in its own context window. A six-agent commit returns in seconds, not minutes — and the bench keeps moving.
III Pixel-art verdict. Each run ends with one of three banners — COUNCIL CLEAN, NEEDS FIXES, or BLOCKER — plus a severity-sorted finding list with file and line citations. VS Code is opened at a click for human inervention. No buried output. Human debugging made easy.
IV Strict-mode secret blocker. A second hook scans staged hunks for credentials — AWS, GitHub, Anthropic, OpenAI, private keys — and refuses the commit before the diff ever leaves the developer computer.
V Permanent findings register. Every review writes to docs/findings_register.yaml — a durable, ALCOA+ trail of what was flagged, by whom, at which SHA. Nothing gets quietly forgotten between sessions. Everything is auditable.
VI Designed for hand-off. The verdict, the citations, and the smoke test all land in the same place the engineer is already looking. Quality review is not a separate ritual — it's the commit.

Spend more time coding, less time in meetings. Better results, faster.

Council verified · every commit six-agent panel · parallel review · permanent register
Open source
{ }
ryogena-pubmed-mcp  ·  A small Model Context Protocol server that lets any MCP-compatible client (Claude Desktop, Claude Code) search PubMed, fetch abstracts, walk the citation graph, and look up an author's papers. Hermetic test suite, zero domain code from Pico — a clean reference for the integration patterns we use internally. MIT licensed.
View on GitHub →
Founder

Kyle Vick

Founder · Ryogena

Fifteen years inside drug discovery R&D as builder and operator. Most recently Associate Director of Research Informatics at SpringWorks Therapeutics, where he founded the AI function and shipped to 94% scientist adoption. Earlier, two and a half years as Lead Solutions Consultant at IDBS / Danaher with #1 regional win rate at top-20 pharma sponsors. Trained as a learning-and-memory neuroscientist at Florida Atlantic University and the Ron Davis lab at The Scripps Research Institute, with peer-reviewed publications in molecular neuroscience and drug discovery. Twenty years coding.