Long story short
From researcher interviews to an on-prem agent pipeline scientists use day to day
01 Discovery
Started from the researchers
Translated 1:1 interviews with 5 researchers into a PRD for the pipeline.
02 Execution
Built the core agent framework
Architected and built the core agent framework: sandboxed execution, custom LLM evaluation, and retrieval across 37 public APIs and a 190K-node public knowledge graph.
03 Adoption
Let usage decide what to unbundle
Shipped an observability UI to track real-world usage. The data showed which tools to unbundle for researchers' daily tasks.
The research is confidential and under publication review, so please email me if you'd like to learn more.