Project Summary
Company
Life Science Venture Fund
Industry
Life Sciences
Services
AI Software Development, AI Agents
Location
London, United Kingdom
A life science venture fund wanted a systematic way to find value in drug programmes the industry had already discontinued. We built Compass, an AI research platform that reviews discontinued programmes, establishes why each one was shelved, and identifies the candidates worth acquiring for a different indication.
Screening at Scale: 827 drugs analysed and 141,000 drug-disease pairs evaluated.
Decision-Ready Output: Every shortlisted candidate comes with a hypothesis card and a proposed trial design.
Continuous Monitoring: Announcements from 100+ pharmaceutical companies are tracked, so new discontinuations surface within hours.
The Challenge
Drug programmes are shelved for many reasons, and not every one of them means the drug failed. For an investor, some of these programmes are acquisition opportunities in a different indication. Finding them requires reviewing each programme, understanding why it was stopped, and judging whether it could succeed elsewhere. That is specialist work, and the fund could not apply it by hand across the full landscape.
The goal was to cover the discontinued landscape systematically, produce output the investment team could act on directly, and to catch new discontinuations as soon as they are announced.
The Approach
We built a pipeline of AI agents that works through each discontinued programme, determines the reason it was shelved, and evaluates the drug against alternative indications. Across the portfolio this meant assessing 141,000 drug-disease pairs.
For each shortlisted candidate, the platform produces a hypothesis card setting out the rationale, together with a proposed trial design, so the investment team starts from a structured case rather than a raw search result.
Alongside the analysis, we built a monitoring service that tracks RSS feeds and SEC filings from more than 100 pharmaceutical companies. The platform runs on a Python and FastAPI backend with a React interface, Supabase for data, and Railway for deployment.
The Impact
The fund now screens discontinued programmes systematically instead of case by case. 827 drugs have been analysed so far, each shortlisted candidate arrives with a hypothesis and trial design ready for review, and new discontinuations reach the team within hours of being announced.
