Work Airstream RAG for Customer Support

Summary

Built an internal support assistant for trailer manufacturer Airstream (Thor Industries) to search across 60GB of manuals and documents.

My Role

AI Lead

Challenges

• Fragmented data across PDF, DOCX, XLSX, and HTML
• Hallucination and irrelevant answers due to poor chunking
• Latency issues due to large corpus and no caching

What We Did

• Cleaned and chunked 60GB+ of technical docs
• Implemented hybrid keyword + RAG with metadata filters
• Integrated query classification and caching

Outcomes

• Latency dropped 3x
• Accuracy improved with internal support teams
• Used internally across product and support