Schematic Spelunker

Problem & user context

Technicians working on specialized machinery rely on dense schematic documents that are hard to search and easy to misread. A wrong interpretation can mean broken equipment or safety risk. Schematic Spelunker (built with a team at a hackathon) turns those schematics into a conversational assistant.

Constraints & tradeoffs

• Hackathon clock. Scope ruthlessly: parse → retrieve → answer, with citations back to the source schematic, before adding anything else.
• Accuracy over fluency. Grounded retrieval (RAG) over freeform generation — an answer that cites the schematic beats a confident guess.

Architecture

Schematic PDFs ──► parser / chunker ──► vector store
                                            │  retrieval
User question ──► LLM (grounded prompt) ◄──┘
                     │
                     ▼
             Answer + schematic citations

Results & lessons

• Working demo shipped within the hackathon window with a 4-person team.
• Lesson: document parsing quality is the ceiling on answer quality — the chunking strategy mattered more than the model choice.