A Privacy-Preserving On-Device Loop: Local LLM Interpretation → FEM Solve → LLM Review for Structural Briefs
by Louie Doniego Balmores & Sandra AgcaoiliLinkedIn · ORCID · ResearchGate · RocketReach · ContactOut · · Balmores Lab
A closed loop that keeps every prompt on the engineer's own machine. A locally-hosted reasoning LLM (DeepSeek-R1 on Ollama, loopback only) canonicalises a plain-English structural brief; the deterministic PyNite finite-element kernel solves it; the same local LLM then reviews the authoritative numeric result and writes recommendations and a conclusion. No prompt, model, or result ever leaves the device. We characterise latency, reasoning-trace scrubbing, loopback security gates, and a deterministic fallback that guarantees the system never fails closed.
privacy-preserving AIon-device inferenceDeepSeek-R1retrieval-augmented engineeringPyNitehuman-in-the-loop
This is a working paper from the Balmores Lab research programme. For credentials and background, see the official profile and CV.
