ConstructGraph — AI Document Intelligence
ConstructGraph lets building owners and project managers upload construction PDFs — RFIs, submittals, change orders, schedule updates — and instantly get structured intelligence. Claude AI classifies each document and extracts entities like costs, dates, spec sections, and responsible parties. A dependency inference engine then detects relationships between documents using pattern matching on those extracted entities — zero extra API calls. The result is an interactive force-directed graph (D3 + React) where you can select any document, enter a hypothetical delay, and watch the impact cascade through connected nodes with weighted decay. All updates stream in real-time via WebSocket.
Try It Live
- Upload a construction PDF and watch AI classify it
- See extracted entities: costs, dates, spec sections
- Explore the interactive dependency graph
- Run impact analysis on a document delay
Key Numbers
Features
- AI document classification (RFI, Submittal, CO, Schedule Update)
- Structured entity extraction via Claude AI
- Zero-cost dependency inference using entity pattern matching
- Interactive D3 force-directed dependency graph
- Impact analysis with weighted delay propagation
- Real-time WebSocket updates on processing events
- AI suggestions with human-in-the-loop review
- S3 presigned URL upload pipeline
- Redis graph cache with automatic invalidation
- 118 backend tests (pytest)
Architecture
┌─────────────┐ ┌──────────────────────────────────────────────┐
│ React SPA │────▶│ FastAPI Backend │
│ (Vite) │◀────│ │
│ │ WS │ Auth ─── Documents ─── Dependencies ─── Graph│
└──────┬───────┘ │ │ │ │ │ │
│ │ JWT S3 Upload NetworkX D3 Data │
│ │ │ │ │ │
│ │ Claude AI PostgreSQL Redis │
│ │ (classify + (persistence) (cache + │
│ │ extract) pub/sub) │
└─────────────┴──────────────────────────────────────────────┘Tech Stack
Key Learnings
Entity pattern matching provides zero-cost dependency inference with deterministic, testable rules — no extra LLM calls needed
D3 force simulation + React SVG hybrid avoids DOM ownership conflicts and lets CSS animations work naturally on graph nodes
WebSocket events triggering TanStack Query cache invalidation is simpler than optimistic updates for backend-driven processing events
PostgreSQL + NetworkX hybrid (DB for persistence, in-memory for algorithms) works well at construction-project scale (<5000 docs)
Want to see more AI projects?
Check out the rest of my AI Lab or get in touch to discuss AI/ML collaboration.