Paid contract. £750+/day. ~2 months (flexible schedule). Weybridge, Surrey / London — hybrid, with the first fortnight on site.
I’ve built an evidence map of ~300 papers on ibogaine, 1957–2026 — a citable research instrument with a DOI on . Schema-governed metadata, Python throughout, a multi-stage extraction pipeline. I’ve carried it a long way on my own. Now I want one excellent engineer beside me to make it hold without me.
Why it has to be right: ibogaine can lift someone out of opioid dependence in a single session. It can also stop their heart. Both of those facts are true, and both live scattered across seven decades of literature — Portuguese theses, autopsy reports, court transcripts, grey material nobody has read. The map exists so a clinician can consult it before they screen someone, and a researcher can cite it and be right to trust it. Which collapses to one requirement: every claim carries its receipt.
That makes this a reliability problem, not an ML one. A real example: one of my validation guards used a pattern that matched nothing — and matching nothing was emitted as a warning, and warnings never fail the build. So the guard sat green on a drifted tree for weeks. “I couldn’t check this” was being scored as “nothing is wrong.” Fail-open where it needed to fail closed. Most of what’s actually load-bearing here has that flavour.
What I need you on:
Coherence — capable machinery in scattered pieces, consolidated into one reproducible pipeline a second person can run end to end.
Reliability by construction — catching fabricated, missing or drifted data before it enters the map. (An AI step once invented a value that looked completely plausible and sat perfectly in range. The errors of omission — the thing that quietly isn’t there — are the harder half.)
Faithful retrieval and provenance — every claim traceable to the page it came from, with a human kept in the loop for the judgement calls rather than automated away.
Built to last — tested, documented, reproducible enough to survive me stepping away for a month.
You: senior, and you think in default-deny, fail-closed, provenance, reproducible builds. Security, identity or regulated-data experience is a strong fit — more so than an LLM CV. You don’t need to know the science; I bring that. I care more about how you think than which frameworks you’ve used.
What you get: £750+/day, no equity and no “exposure” — and a named contributor credit on the citable record. CITATION.cff and the .
I read every reply properly, and I’d rather you tell me how you’d shape this than have me hand you a rigid spec.
Questions or applications: philip[at]ibogaine.space