Skip to Content
Back to Portfolio

BBS Harness — AI Structural Drawing Analyzer

Ongoing1 developer

Project Under NDA

Visuals not available for confidential projects

A multi-stage LangGraph agentic pipeline that extracts bar bending schedules from structural engineering PDFs, using durable Postgres-backed checkpointing for resumable, crash-resilient orchestration with human-in-the-loop review gates.

Technologies Used

PythonLangGraphComputer VisionPostgreSQLLLM Agents

The Challenge

Structural engineers manually extract bar bending schedule data from complex PDF drawings — a tedious, error-prone process that creates bottlenecks in construction planning.

The Solution

Architected a multi-stage agentic pipeline with human-in-the-loop review gates, multi-modal progressive inspection, and intelligent token budget management to automate accurate data extraction from structural PDFs.

Key Features

Multi-stage LangGraph pipeline with Postgres-backed durable checkpointing
Human-in-the-loop (HITL) review gates with typed interrupt boundaries
Multi-modal progressive inspection: low-res overview → region detection → detail crop extraction
Intelligent token budget management with payload estimation and message pruning
Evidence linking for extracted values back to source regions

Impact & Results

Automates the extraction of bar bending schedules from structural drawings, reducing manual data entry time from hours to minutes while maintaining engineer oversight through review gates.

My Role: AI Engineer

Interested in working together?

Send me an email (ayesigwar@gmail.com)