AeroForge
AI-powered production risk intelligence for aerospace manufacturing. A research-driven prototype designed to help manufacturing teams identify potential production delays before they become costly operational problems.
Production risk is often visible in pieces.
Aerospace manufacturing involves complex interactions between suppliers, components, production stages, equipment, inspections, rework and workforce capacity.
The product opportunity I explored was simple: How might manufacturing teams identify emerging production risks early enough to investigate them?
AeroForge is an independent research and prototyping project. It does not use confidential aerospace data and is not intended for safety-critical decision-making.
Designed around the people making production decisions.
Production Manager
Primary user responsible for monitoring production progress and identifying areas requiring intervention.
Quality Manager
Needs visibility into defects, inspection failures and rework signals that may affect production.
Supply Chain Manager
Needs visibility into supplier performance, component shortages and delivery-related risks.
From fragmented signals to one risk view.
Product hypothesis
If production, supplier, quality and operational signals are brought together and converted into transparent risk estimates, manufacturing teams may be able to identify areas requiring investigation earlier.
Core user story
As a production manager, I want to identify production units with elevated delay risk so that I can investigate contributing operational factors earlier.
Five capabilities form the first version.
01
Production Overview
High-level visibility into production volume, delay patterns and operational signals.
02
Risk Prediction
Machine-learning predictions estimate the probability that a production unit may experience delay.
03
Supplier Intelligence
Surfaces supplier-related signals associated with production delays.
04
Quality Intelligence
Connects defects, failed inspections and rework with production risk.
05
AI-Assisted Explanation
Uses an LLM to translate model evidence into human-readable operational insights.
A synthetic manufacturing dataset.
Because this is an independent portfolio project, I created a synthetic dataset designed to represent production-level signals that could be relevant to manufacturing risk analysis.
Signals represented
- Supplier delivery performance
- Supplier delay days
- Component shortages
- Machine downtime
- Defect counts
- Failed inspections
- Rework hours
- Workforce load
- Planned duration
- Actual production duration
Predicting production delay risk.
I built a classification pipeline using Logistic Regression. Categorical variables are one-hot encoded while numerical variables are standardized through a preprocessing pipeline.
The target variable is production_delayed. The dataset was split into training and testing sets using a stratified 80/20 split.
Important: these performance metrics are calculated on synthetic data created for this prototype. They should not be interpreted as evidence of performance on real aerospace manufacturing operations.
Predictions become operationally understandable.
Rather than displaying only a probability score, AeroForge translates model output into three risk categories.
| Risk Level | Probability | Interpretation |
|---|---|---|
| LOW | < 45% | Low predicted likelihood of production delay. |
| MEDIUM | 45% – 74% | Monitor contributing production signals. |
| HIGH | ≥ 75% | Investigation recommended. |
The model predicts. AI helps explain.
Machine Learning
The ML model calculates the probability of production delay from structured production signals.
Generative AI
Groq-powered LLM capabilities convert structured model evidence and identified risk factors into concise, human-readable explanations.
The intelligence is exposed as a service.
AeroForge also includes a FastAPI backend, separating the intelligence layer from the dashboard interface and creating a foundation for future integrations.
FastAPI
Backend API layer for serving AeroForge intelligence.
Swagger
Interactive API documentation used to verify backend functionality.
Future Integrations
The API architecture creates a path toward integration with manufacturing applications and enterprise systems.
A decision-support dashboard.
The Streamlit prototype brings the intelligence together into a single interface for exploring production risk, supplier performance, quality signals and model performance.
Production Risk
Identify production units with elevated predicted delay probability.
Supplier Intelligence
Explore supplier-level patterns associated with delay.
Quality Intelligence
Explore relationships between defects, inspections, rework and delay.
Model Performance
Review accuracy, precision, recall, F1 and ROC-AUC.
AI-Assisted Insight
Receive a plain-language explanation of identified production risk factors.
The dashboard is a prototype for demonstrating product thinking, data analysis and AI integration. It is not a certified aerospace manufacturing system.
The value is earlier visibility.
For Production Teams
- Prioritize units requiring investigation.
- Move from reactive reporting toward proactive monitoring.
- Bring multiple operational signals into one view.
For Management
- Improve visibility into production health.
- Identify recurring operational patterns.
- Create a foundation for future predictive operations.
This wasn't just a model.
AeroForge demonstrates my approach to AI product development: start with a real operational problem, define the user, understand the data, prototype the intelligence, expose it through an API, and turn the output into something a decision maker can actually use.
Product
Problem framing, user definition, MVP scope, product hypothesis and business value.
Data
Synthetic data generation, validation, feature engineering and exploratory analysis.
AI Engineering
Machine learning, risk scoring, explainability, generative AI, API development and testing.
Designed with clear boundaries.
What AeroForge does
It demonstrates how structured manufacturing signals could be used to create an experimental production-risk intelligence workflow.
What AeroForge does not do
It does not control equipment, make autonomous production decisions, replace manufacturing experts, certify aircraft, or make safety-critical decisions.
Any future production implementation would require representative operational data, rigorous validation, domain-expert review, cybersecurity controls, model monitoring and appropriate regulatory and organizational governance.
Built across the product and technical stack.
Programming
Python · Pandas · NumPy
Machine Learning
Scikit-learn · Logistic Regression · Feature Engineering
AI
Groq API · Generative AI · AI-assisted explanations
Backend
FastAPI · Uvicorn · REST API
Frontend / Prototype
Streamlit · Plotly
Quality
Pytest · Model evaluation · Data validation
From prototype to production research.
The next stage would focus on validating the concept against real-world manufacturing workflows rather than simply improving the model score.
Real Data
Evaluate the approach using representative, appropriately governed manufacturing data.
Model Validation
Test calibration, generalization, drift and performance across different production environments.
Human-in-the-Loop
Study whether risk explanations actually help production teams make better investigation and prioritization decisions.