Designed and built a production-ready

Designed and built a production-ready internal AI support bot from scratch. The system uses a five-layer RAG architecture to ingest and retrieve knowledge from different knowledge sources, e.g. Confluence, Jira, Notion, and Slack, enabling staff to get accurate, source-cited answers to internal queries directly within Slack.

Built entirely on internal infrastructure using the Claude API, Supabase, and pgvector - ensuring zero data exposure outside the organisation. The knowledge base spans 10,000+ entries across 4 data sources, with hybrid retrieval combining semantic search, keyword fallback, and query rewriting for accuracy.

The bot replaced a manual support model, reduced repetitive query handling, and introduced a scalable self-service layer for internal IT support / helpdesk - while resolving the data privacy concerns that had blocked a third-party vendor solution.