Inspiration
CAE engineers can spend hours reviewing frequency-response curves, locating critical peaks, comparing multiple directions against targets, and translating results into decision-ready reports. The calculations are repeatable, but interpreting the evidence and choosing the next engineering action still requires specialist judgement.
CAE Insight Copilot was created to shorten that path from simulation results to engineering action.
What it does
CAE Insight Copilot provides an interactive workspace for vibration transfer-function analysis. A user can upload frequency-response CSV data or use the included synthetic demonstration dataset.
The application:
- Plots X, Y and Z response curves against an adjustable engineering target.
- Detects and ranks governing exceedances and near-limit watch events.
- Identifies the critical frequency, direction and maximum response.
- Generates an evidence-grounded GPT-5.6 engineering assessment.
- Separates measured evidence from hypotheses that require confirmation.
- Recommends investigation steps, countermeasure studies and a validation plan.
- Exports the assessment report and demonstration data.
The deterministic rules engine performs the numerical checks. GPT-5.6 then receives structured evidence and turns it into a clear engineering narrative without replacing engineering judgement.
How I built it
I used Codex as my development partner to accelerate product planning, Streamlit implementation, deterministic peak-ranking logic, visual design, OpenAI Responses API integration, documentation, testing and deployment troubleshooting.
The application is written in Python using Streamlit, Pandas, NumPy and Plotly. GPT-5.6 is called through the OpenAI Responses API. The API key is stored securely as a Streamlit deployment secret and is never displayed or committed to the public repository.
Challenges
The main challenge was combining deterministic engineering calculations with generative interpretation without allowing the model to invent unsupported conclusions. I addressed this by calculating target exceedances, peak rankings and watch events before the model call, and then supplying GPT-5.6 with structured evidence and explicit instructions to distinguish facts from hypotheses.
Other challenges included creating a readable engineering dashboard, protecting API credentials, resolving local Windows execution issues, and deploying a reproducible public demonstration.
Accomplishments
The completed project provides a coherent end-to-end workflow: data input, interactive visualisation, deterministic compliance checking, GPT-5.6 interpretation, recommended actions, validation planning and downloadable reports.
The public application and repository allow judges to test the project without rebuilding it locally.
What I learned
I learned how Codex can support the full development lifecycle rather than only generating isolated code. I also learned that AI is most useful in engineering decision support when deterministic calculations establish the evidence and the model focuses on communication, prioritisation and next-step reasoning.
What's next
Future versions could support additional CAE result formats, automated modal correlation, multiple response locations, project-to-project comparisons, tolerance studies and integration with enterprise simulation workflows.
Log in or sign up for Devpost to join the conversation.