Senior Full-stack engineer

Riffbot Context Pipeline - Hybrid RAG for AI Chat for an AI-powered creative technology company – Built a hybrid RAG system to bring targeted context into the Riffbot chatbot, improving response relevance by retrieving the most useful help docs and user content (pins) for each conversation. – Engineered hybrid retrieval combining pgvector semantic search (HNSW-indexed, 768-dim Gemini embeddings) with PostgreSQL full-text search, merged via Reciprocal Rank Fusion with custom BM25 re-ranking in application code (Cloud SQL limitation). – Designed LLM-based semantic chunking using Gemini Flash to group content by topic, with intelligent fallbacks for short content and LLM failures. Content hash deduplication skips re-embedding unchanged documents. – Built async Pub/Sub pipeline for embedding tasks with graceful fallback to direct execution in dev, daily cron sync for help docs via Firecrawl scraping, and admin tooling for manual triggers. Teams Platform for an AI-powered creative platform – Owned green-field development of a collaborative workspace product from concept to launch, serving 100+ creative teams across organisations up to 100 members. – Engineered multi-tenant architecture with team-based asset isolation, implementing a custom teamId solution to work around GraphQL header limitations in Hasura. – Built a batched asset transfer system with optimised FK constraint handling to prevent database locks during team member transitions.