Freelance AI Engineer for Biology & TechBio Teams
I’m Vivien Perrelle, an AI engineer specialized in AI for biology. I want to work on systems that genuinely move science forward, and I believe the bottleneck now is trust: AI that is plausible is not enough when the output has to be correct. So I build the verification and reproducibility layer that scientific AI needs. I focus on AI agents, context engineering over scientific data, evaluation harnesses, and claim-to-source verification for research workflows.
What I do
My work sits where language models meet real scientific workflows, the part that has to be right rather than just plausible:
- AI agents for research workflows: agents that read, extract, cross-check, and draft over your internal corpus, with humans approving the irreversible steps.
- RAG over scientific literature and internal data: retrieval pipelines grounded in primary sources, built to cite what they claim.
- Claim-to-source verification: deterministic checking that every statement in a generated or human document traces back to its evidence.
- Evaluation harnesses: measurable baselines and regression suites so you know whether the system actually improved.
- Scientific data tooling: pipelines over CrossRef, OpenAlex, PubMed, and your own datasets.
Who I work with
I want to work with seed-to-growth TechBio and AI-for-biology startups, and AI-for-science teams inside larger organisations, at the moment a demo has to become a dependable system. That transition is what I’m built for: an AI for biology engineer who writes production code, not slide decks.
Why me
I’ve worked on both sides of the problem, the biology and the software:
- Built a smartwatch with embedded enzymatic biosensors during my MSc in Creative Technologies at the De Vinci Innovation Center.
- Hands-on R&D at PKvitality, a VC-backed team building a non-invasive CGM smartwatch.
- Founder of LocusLab: independent evidence-assurance infrastructure for biology and regulated science.
- Author of an open-source scientific claim verifier reaching F1 0.92 on SciFact (vs 0.62 naive baseline).
- Shipped production AI agents in my own startups, Finexov and Oseille AI.
How I work
I fit in well with small startup teams. I go all-in on one problem at a time. I work remotely on CET hours, with comfortable overlap for EU and US-East teams, and I adapt to your team’s rhythm. I’m also glad to relocate for long-term missions or roles.
Read my thinking
- AI for Science Is Moving From Prediction to Closed-Loop Research Systems
- Science Is Entering Its Agentic Era
- Regulators Don’t Accept Vibes: The Two Layers Pharma AI Is Missing
Common questions
Do you work with early-stage TechBio startups?
That’s exactly the work I’m set up for. Early teams usually need one system taken from prototype to production fast, with verification built in so it holds up. I can take that on as a defined project, or embed with the team that owns it.
What stack do you work in?
Python (async, FastAPI, Pydantic), the major LLM APIs and agent frameworks, retrieval infrastructure, and Docker-based deployment, including on-prem and air-gapped environments when the data can’t leave your infrastructure.
How do engagements start?
A 30-minute intro call. If the problem is a fit, I send a concrete proposal within a few days: scope, deliverables, timeline. If it isn’t, I’ll say so directly.