Skip to content

Problem statement and positioning

Start here. Primary project document for reviewers and interviews.
Disclaimer: disclaimer.md — unofficial portfolio demo.


Core positioning

CAE Copilot is a documentation assistant for CAE software workflows — not a solver replacement and not a live automation bridge to vendor GUIs.

The product is designed as a multi-CAE platform with pluggable adapters. The first adapter targets ANSYS Rocky (DEM) using official v26r1 manuals indexed locally with citations.

One-line summary:

CAE documentation copilot — cited answers from Rocky manuals, tutorial steps, script drafts, and deterministic DEM helpers. Human engineer stays in control.


The problem (Rocky engineer)

ANSYS Rocky ships extensive documentation: user manual, tutorial guide, scripts API, module reference, CFD/SPH technical manuals, verification guide. Engineers spend significant time:

  • finding the right section across multiple PDFs,
  • translating tutorial prose into repeatable setup steps,
  • looking up Python API patterns for app.GetStudy() and particle processes,
  • cross-checking coupling options (Fluent, one-way vs two-way).

CAE Copilot reduces navigation and retrieval time while keeping every claim traceable to a manual page.


Five engineering problem areas

These are the enterprise adoption failures the platform design addresses. The Rocky adapter is the first concrete demo of each response:

# Problem area Rocky demo response
1 Document chaos — thousands of PDF pages, no governed index Manifest-driven ingest, page-aware chunks, /kb/stats
2 Black-box AI — answers without auditable sources Mandatory citations, eval citation hit rate
3 AS-IS automation — GUI macros on broken workflows Documentation-first; live solver control deferred (Stage 3)
4 Vendor / model lock-in — single cloud LLM or CAE stack Model-agnostic LLM; pluggable adapter per CAE product
5 Unvalidated calculations — arithmetic inside the LLM Deterministic DEM tools with Pydantic validation + retry

See Enterprise patterns below for the portfolio framing table.


What this demo does (and does not)

In scope Out of scope
RAG over Rocky v26r1 PDFs with page citations Live Rocky COM/API control
Tutorial-style setup answers Running DEM jobs from the copilot
Script drafts for manual copy into Rocky Publishing Ansys docs in git/images
DEM helper tools (particle count, units, timestep heuristic) Replacing Ansys support

Demo paths

Knowledge:   Question → RAG (manuals) → Cited report
Setup:       Tutorial question → RAG → Numbered steps
Scripting:   API question → RAG → Draft .py (copy to Rocky)
Calculation: Numeric helper → Tool → Report

Enterprise patterns (portfolio framing)

Pattern CAE Copilot response
Black-box AI Mandatory citations + eval golden set
Vendor lock-in Model-agnostic LLM; adapter per CAE product
GUI macro automation Documentation-first; live integration deferred

Stage 3 in the roadmap is live CAE integration (license/API required) — honestly deferred for this portfolio build.

See walkthrough.md for the 3-minute demo script.