Technical lead for an early-stage AI platform that automates investment research workflows, including expert interview calls, transcription, and research knowledge management.
- Drove the team's migration of the legacy workflow-engine backend to a TypeScript agent service, exposing existing workflows as agent tools and standing up CI and test gates alongside.
- Designed the split of the backend into a shared cloud service and a local service behind a clean API boundary, so sensitive research data stays on the customer's machine by default, and built most of the local service.
- Co-built the knowledge-management backend, where agents read and write the user's files, answer research queries over them, and use vector search with LLM inference, owning the ingestion connectors for meeting recorders, multiple Google accounts, and network storage.
- Created a shared tool layer for AI agents and exposed it over MCP and an OpenAI-compatible API, so the native app, third-party chat UIs, and coding agents reuse the same backend capabilities without duplicating logic.
- Designed financial data pipelines that combine deterministic processing with LLM-assisted extraction, validation, and enrichment.
- Built backend services for automated call and transcription workflows, with a focus on reliability, operational visibility, and safe external integrations.
- Reviewed system designs, wrote risk analyses, and called out critical architecture decisions early so teams could avoid costly rework.