SpecCheck: Intelligent Extraction System for Construction Bidding Documents
From 72% baseline to 87% accuracy on 29 scattered bid fields — with highlighted PDF proof
A Django-based application that accepts construction bidding PDFs up to 500MB and automatically extracts 29 structured data fields using an advanced multi-phase RAG pipeline.
Built at 1POINT1 for construction estimators who need bid deadlines, bond amounts, and scope fields transcribed accurately — not guessed from a single LLM skim.
- Role
- AI Engineer at 1POINT1
- Latency
- ~70–80 seconds per document
- Accuracy
- 81–87% overall, ~92% critical fields
- Duration
- 5-phase accuracy optimization program
- Team Size
- Engineering delivery team