MemInfra
Built a reusable AI memory infrastructure layer that enables LLM applications to extract, store, update, and retrieve long-term user and conversation knowledge. The system supports semantic and episodic memories, deterministic deduplication and supersession, hybrid lexical + vector retrieval, importance/recency-aware ranking, conversation summaries, and per-user memory isolation. Designed the memory lifecycle from LLM-based extraction through policy-driven validation and persistence, with FAISS-based semantic search and evaluation tooling to measure retrieval quality across memory benchmarks.
PythonPostgreSQLSQLAlchemyFAISS
Agentic Intelligence Platform
Built an enterprise agentic AI platform that combines LLM reasoning with deterministic validation for reliable workflow automation. Implemented tool orchestration, DAG execution, evidence-grounded reasoning, RAG, evals, observability, provenance, execution budgets, and governed actions with human approval.
PythonFastAPIPostgreSQLSQLAlchemy
perps-trading
Built a perpetual futures exchange backend with a deterministic matching engine supporting limit/market orders, leveraged long/short positions, margin management, fills, liquidation-price calculation, and real-time order-book updates. Designed an event-driven architecture using Redis Streams, REST APIs, WebSockets, and Binance market-price feeds.
Node.jsTypeScriptRedis StreamsWebSockets