case-study

Schematic Spelunker

AI chatbot that parses machinery schematics so technicians get informed answers about their unique equipment.

  • React
  • Rust
  • Gemini API
  • RAG
  • Document Parsing
Schematic Spelunker cover

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 was parse, retrieve, and answer, with citations back to the source schematic.
  • Answers cite the source schematic. Grounded retrieval (RAG) limits freeform generation.

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.
  • The chunking strategy affected answer quality more than the model choice.