Structural engineer's AI workflow: LLMs audit divergences, humans sign off
Kooky-Sorbet-5996 · reddit · 2026-11-06
A structural engineer shares a three-layer workflow for AI-assisted steel structure code checks, arguing LLMs are great at surfacing divergences but not at deciding:
- Deterministic layer: PDF extraction (regex/OCR), unit normalization (kNcm→kNm, MPa→kN/cm²), and numerical formula closure done in Python—not the LLM—plus sanity checks (η≤100%, MSd≤MRd, Δ≤30%).
- Chat layer: AI audits assumptions (is Cb forced to 1.0? is kv=5.34 valid?), flags code conflicts (EC3 Annex B ≠ NBR 5.5.1.2), classifies divergences, and builds a decision queue.
- HITL layer: the engineer validates premises, picks the methodology, and signs the result.
Key takeaway: keep computation deterministic, use LLMs for conceptual auditing, and reserve final engineering judgment for humans.
More from coding & agent
- Sentry CEO shares screenshot: one AI coding agent spent $2,867 in 7 days — zeeg · 2026-11-06
- LangChain shares mental models for high-quality LLM evals from harmonic's Austin Berke — LangChain · 2026-11-06
- GraphRail: graph-grounded step-level verification catches search agent reasoning errors without an LLM judge — _reachsumit · 2026-11-06
- Weaver: Incremental Reasoning Over Conversation Memory for Personalized RAG — _reachsumit · 2026-11-06
- VoidZero ships tsrs Rust TypeScript port (8x faster), open-sources Void, joins Cloudflare — cnakazawa · 2026-11-06
- 35 of 68 Codex Plugins Unused in Two Weeks: Idle Plugins Bloat Your Context Window — nickbaumann_ · 2026-11-06