Founding Engineer
Axy Inc · Aug 2025 – Present
- Joined as the first engineer hired and helped rebuild the product from an MVP into a production scientific knowledge graph platform; co-designed the Turborepo monorepo, BFF/API gateway, service boundaries, shared packages, and backend architecture while working closely with the CTO, designer, and scientists on product requirements.
- Built and scaled data systems supporting millions of scientific knowledge graph entities across PostgreSQL, MongoDB, Qdrant, and Redis, powering graph exploration, research workflows, document management, and semantic search.
- Designed and implemented the end-to-end submission workflow for researchers to draft graph changes, review and approve submissions, detect conflicts, maintain mutation history, and safely merge changes into the main graph using a delta-based model.
- Moved cross-service synchronization and long-running submission merges out of synchronous API requests using AWS Lambda, SNS, and SQS; added dead-letter handling and reconciliation checks to keep graph, search, vector, and document services consistent.
- Built a multi-stage retrieval pipeline using dense and sparse search in Qdrant, Reciprocal Rank Fusion, and reranking, improving retrieval quality from roughly 0.5 baseline to 0.7-0.8 across scientific BEIR evaluations using Recall@K, NDCG@10, and MRR.
- Owned major parts of the applied-AI pipeline including embeddings, hybrid retrieval, query rewriting, reranking, context selection, agent/tool workflows, chunking, and evaluation; also built a Python prototype for converting scientific papers into structured knowledge graph entities.
