Energy & Technology Company Harnesses AI to Revolutionize Rebar Detailing in 8 Weeks

Founder of an energy & technology company, envisioned an AI solution to automate rebar detailing—a labor-intensive bottleneck in structural engineering. Detailing transforms architectural plans into construction drawings, typically requiring 20+ hours of manual work per structure.

This energy and technology company set out to prove that AI could fully automate rebar detailing for drilled piers in just 8 weeks. The goal was end-to-end automation—from drawing upload to construction-ready output—with automatic validation against ACI 318-19 and CRSI standards. Beyond compliance, the solution needed to extract complex specifications, generate accurate cross-sections and elevation drawings, and establish commercial viability for expansion across additional structure types.

Rebar detailing is one of engineering's most demanding disciplines. Engineers must interpret ambiguous architectural drawings, cross-reference geotech reports and bar schedules, and apply hundreds of pages of ACI/CRSI code requirements—all with zero margin for error.

‍

The calculations alone are complex: variable spacing zones, lap splices, development lengths, and clearance covers, while distinguishing per-structure from total quantities and producing industry-standard CAD outputs.

‍

For AI to take this on, it had to master multimodal synthesis across drawing types, maintain absolute accuracy with no tolerance for hallucination, and work within the constraints of AWS Bedrock's architecture—all while inferring three-dimensional structures from two-dimensional drawings.

‍

Protagona built a multi-agent AI architecture on AWS Bedrock, consolidating six specialized agents into a single Civil Engineer Agent—delivering superior performance with a leaner, more maintainable system.

‍

At the core is a hierarchical knowledge base trained on ACI 318-19 and CRSI standards, enabling context-aware code interpretation that fixed-token chunking simply couldn't achieve. Advanced prompt engineering reduced hallucinations by 90%, allowing the AI to accurately interpret complex spacing zones, distinguish per-pier from total quantities, and cross-reference values across multiple drawing sheets.

‍

Construction-ready cross-section and elevation drawings are generated automatically via Python automation—a breakthrough delivered in days rather than the planned weeks, with 100% accuracy. An iterative validation loop ensures every output meets code requirements before delivery, automatically correcting non-compliant designs and flagging unresolvable issues for human review.

‍

Two additional innovations made it production-ready: a dynamic tiling strategy that overcomes AWS Bedrock's image processing limits without losing drawing context, and a missing data detection system that prompts users for incomplete specifications rather than fabricating values.

‍