Biocompile is building an AI co-scientist platform for biotech R&D. Their AI agent already finds what researchers ask for. The opportunity was to go further: surface the connections no one had made yet. Our challenge was to make it intelligent enough to discover what nobody thought to connect.
Biocompile's platform structures research around a unified data model: papers, experiments, hypotheses, pipelines, and datasets. The problem isn't finding information, it's surfacing the connections no one made: two experiments on related genes that belong together but were never linked. As Biocompile scales, this limitation compounds.
Graph-Native Data Layer
AWS Neptune Analytics combines a graph database and a vector store in a single serverless layer. The entire research corpus is ingested as a typed property graph, with vector embeddings attached to each node. Typed edges encode semantic relationships.
Three-Tool Query Pipeline
Semantic search, graph exploration, and entity reading. A query begins with vector search; those chunks then anchor a graph traversal, pulling in semantically related entities the vector search alone would have missed. Results are re-ranked before passing to Claude Sonnet via Bedrock for synthesis.
Evaluation Framework
DeepEval benchmarking against a curated golden dataset of representative research queries.
Working with Apex was a great experience. Their team was competent and proactive, communication was fast and frictionless, and they genuinely seemed to care about the project. Apex clearly defined the scoped of the project and provided updates regularly which made progress throughout transparent. Within the defined period they fully delivered on the set out goal.
Aaron Riedling CTO
The Research Sub-Agent validated the core thesis: graph traversal surfaces connections that pure vector search cannot find. For Biocompile, this closes a real capability gap in research discovery across teams. The next step is integrating the sub-agent into the core AI co-scientist platform.
Tell us more about yourself and what you're got in mind.
Prefer email? hello@apexlab.io
Or even a quick call?
We don’t care about fancy CVs or long interview processes. Provide some basic information, let’s have a quick chat, and then show us what you love doing the most: write some code and create a product.
