Giao Le Xuan

• Designed, built, and shipped Bridz end-to-end: a multi-tenant customer-feedback platform (Laravel + Vue) replacing Canny across Qikify and Teeinblue (~3,000+ engaged users).
• Built its core platform features: a vote-velocity trending algorithm, Brevo-backed notifications, and SSE-based real-time task streaming over Redis BLPOP.
• Built a centralized Typesense search service indexing ~2–3M entities, owning the indexing pipeline and query/filter design.
• Built a generic soft-delete / restore layer across nearly all customer-facing models, eliminating accidental-deletion data loss.
• Implemented rate limiting on Shopify-storefront endpoints, using encrypted-header anonymous identity to harden anti-abuse.

• Built and maintained backend services in ExpressJS / TypeScript / MongoDB.
• Designed REST APIs from product requirements.

Open-source local-first long-term memory for LLM agents

• Local-first memory layer that any LLM reads and writes over MCP, so context survives across sessions and tools instead of resetting each conversation.
• Model-agnostic hybrid retrieval over BM25, semantic vectors, and knowledge-graph search, re-ranked with a local cross-encoder; runs on-device with no external API key.
• Designed and shipped solo, end-to-end: retrieval engine, MCP server, ingestion pipeline, and CLI, published to PyPI under MIT.
• Used daily as my own working memory.