Project overview
AIQ Product Guide was designed as an intelligent librarian for product knowledge inside Slack. The platform ingests support documentation, the public website, API docs, Slack conversations stored in Snowflake, and media with OCR/transcription, generating atomic Q&A pairs when appropriate. At query time, the system retrieves evidence across multiple vector indexes, searches code through Zoekt when relevant, reranks sources, synthesizes the answer with OpenAI, evaluates correctness, and assembles a response with citations, links, and confidence.
Challenge
Reduce time spent by product and support teams searching scattered information while keeping answers reliable across heterogeneous sources, long conversations, attachments, ambiguous terms, stale information, and hallucination risk.
Solution
Built a Python architecture with LlamaIndex, OpenAI, Vertex AI Vector Search with local fallback, Snowflake for state and Q&A, domain-specific crawlers, PII redaction, media processing, Zoekt code search, correctness scoring, iterative refine/retrieve, and Slack ops commands such as /ops-ingest, /ops-retry, /ops-index, and /ops-status.
Tech Stack
- Python
- OpenAI
- LlamaIndex
- Vertex AI
- Snowflake
- RAG
Technical scope
- Slack, web docs, API docs, and media ingestion
- Vertex AI Vector Search with local fallback
- Cited synthesis with confidence and correctness scoring
- Slack operations for retries, status, and reindexing