An agentic AI platform for biotech research — in production, in daily professional use. Otis is a pseudonym; the public examples are blinded.
Otis plans its own research, reads what the world knows and what the firm knows, and cites every claim it makes.
Biotech research is a breadth problem. The story of any one company is scattered across SEC filings, clinical-trial registries, and a steady drip of news — and it changes weekly. Staying genuinely current on dozens of companies at once is a full-time job for a team. Otis makes it an automated one.
At the center is a glass-box research orchestrator. Given a company, it plans its own research, executes that plan in parallel across SEC filings, clinical-trial registries, and news and web sources, and synthesizes the findings into a report in which every claim carries a citation back to its source. Glass-box means exactly that: no step is a black box — the plan, the queries, and the sources are all visible, logged, and testable.
The public record is only half of it. Otis was built for a biotech-focused advisory firm, and what makes the system genuinely powerful is that it pairs the external evidence with the firm's own internal knowledge — its document libraries and its email. What the firm's researchers already know about a company, and how they plan to engage it, used to live scattered across folders and inboxes; Otis reads it alongside the filings and the news, so a report reflects not just what the world knows, but what the firm knows.
On top of the reports sits an assistant that answers questions conversationally — and then double-checks its own work. A second AI pass reviews each answer against the underlying sources, because in professional research, confidently wrong is worse than slow.
Otis runs in production on Azure, where autonomous scheduled jobs keep its research fresh without anyone pressing a button. It has been in daily professional use since early in its build.
Confident answers are easy to generate. The harder and more useful thing is an answer that checks itself against the sources before it reaches a user.
Given a company, the orchestrator drafts its own research plan.
Executes in parallel across filings, registries, news, and the web.
Writes a citation-backed report; every claim traces to a source.
A second AI pass reviews answers before they're trusted.