Francis Burnet AI Engineering Portfolio
A production-facing portfolio that transforms class capstones into live, reviewable AI workflows across data science, machine learning, and deep learning.
Built From Scratch — Full-Stack AI Portfolio
Every layer of this site was designed and engineered custom — from the server infrastructure and PHP application layer to the front-end styling, AI model hosting, and CI/CD-style deploy pipeline. There is no CMS, no page builder, and no third-party template.
Server & Hosting
- Plesk on Linux VPS
- PHP 8.x via FastCGI
- SFTP + SSH deploy pipeline
- Laravel Herd (local dev)
Back-End
- PHP — custom MVC-style routing
- JSON-driven result feeds
- Dynamic includes architecture
- Per-capstone content modules
Front-End & Branding
- Bootstrap 5 (custom-configured)
- Hand-authored CSS design system
- Custom FB logo + monogram
- Responsive across all breakpoints
AI & ML Layer
- Python / TensorFlow / Keras
- TensorFlow.js (in-browser inference)
- Teachable Machine integration
- Jupyter notebooks (Colab-ready)
Data & Storage
- Git LFS for large datasets
- CSV / image / JSON datasets
- GitHub as source-of-truth
- Exported model artefacts on-server
Deploy & DevOps
- PowerShell build + zip pipeline
- Posh-SSH SFTP upload
- SSH unzip-to-docroot on server
- Git versioned every release
Notebooks & Science
- pandas, NumPy, scikit-learn
- Matplotlib / Seaborn
- EfficientNet, ResNet50, LSTM
- K-Means, PCA, Market Basket
Toolchain
- VS Code + GitHub Copilot
- Google Colab (GPU training)
- Python venv (3.12, local CPU)
- PowerShell 7 + Posh-SSH
Training Highlights
Each capstone runs a real training pipeline — raw data, cleaned splits, model fit, evaluation, and exported artefacts. Below are the standout projects.
Autonomous Driving Vehicle Detection
Published dual-track project combining YOLOv8n vehicle detection, baseline model comparison, and a Tesla autopilot fatality analysis lane. The notebook preserves the full trial-and-error path instead of hiding weaker earlier attempts.
Vessel Type Classifier for Port Operations
Published port-operations project comparing a custom CNN against transfer learning, preserving the metrics, training curves, and final recommendation alongside a browser-based demo.
Landsc8pe Instant Lawn Quote Concept
Standalone live-class concept page that turns a yard photo into a lawn-service quote, launch promotion, and booking flow with a more product-style presentation than the capstone pages.
3-Class Face Mask Detector
EfficientNetB0 and ResNet50 trained on 12,000+ images across three classes: with_mask, without_mask, mask_worn_incorrect. Model exported to TensorFlow.js and Teachable Machine — both run live in-browser.
Bike Rental Demand Forecasting
Regression pipeline on UCI Bike Sharing dataset: feature engineering on weather/season/time variables, Random Forest + Linear Regression comparison, RMSE and R² evaluation.
Adult Census Income Classifier
Binary classification on the UCI Adult Census dataset: encode + scale, train Decision Tree / Logistic Regression / Random Forest, ROC-AUC comparison, confusion matrix analysis.
Mall Customer Segmentation
K-Means clustering on spending-score and income features. Elbow-method optimal-k selection, PCA 2D visualisation of cluster centroids, actionable segment labels.
Market Basket Analysis
Apriori algorithm on 7,500 retail transactions. Frequent itemset mining, confidence/lift-ranked association rules, actionable product-pairing insights.
MNIST Digit Recognition (CNN)
Convolutional Neural Network on 70,000 handwritten digits. Multiple conv/pool/dense configurations compared, training curves logged, test accuracy benchmarked against baseline MLP.
Datasets Hosted on GitHub
Browse repo ↗All datasets are version-controlled directly in the GitHub repo alongside their notebooks. Image data uses Git LFS — notebook clones pull full binary data automatically, no manual download step.
| Dataset | Rows / Size | Task | Capstone | |
|---|---|---|---|---|
| NSMES1988 (Health Survey) | Multi-year survey · health + socioeconomic | EDA / Regression / Classification | Sessions 1–4 — Applied Data Science | ↗ |
| Florida Bike Rentals | 17,379 rows · weather + time features | Regression (Random Forest / Linear) | Session 5 — Machine Learning | ↗ |
| Adult Census Income | 48,842 rows · 14 features | Binary Classification (DT / LR / RF) | Session 6 — Machine Learning | ↗ |
| Credit Card (CC General) | Transactional · spending features | Clustering / Unsupervised (K-Means, PCA) | Session 7 — Machine Learning | ↗ |
| Movies & Ratings | User–item rating matrix | Collaborative Filtering / Recommendation | Session 8 — Machine Learning | ↗ |
| Churn Modeling | 10,000 rows · customer demographics + account | Binary Classification (ANN) | Session 9 — Deep Learning | ↗ |
| Face Mask Images | ~12,000 images · 3 classes · Git LFS | Image Classification (EfficientNetB0 / ResNet50) | Session 10 — Deep Learning | ↗ |
| Grammar & Product Reviews | Text corpus · product review sentences | Sentiment Analysis (LSTM / NLP) | Session 11 — Deep Learning | ↗ |
Interactive Playgrounds
Explore the models and tools that power this portfolio — directly in your browser, no install required.
Autonomous Driving Demo Viewer
Open the published autonomous driving project viewer to compare baseline bounding-box outputs, the final YOLO grid, and the Tesla autopilot analysis visuals in one browser-friendly page.
Open Driving DemoFace Mask Detector — Live
Upload an image or use your webcam to run the 3-class ResNet50 capstone model in real time — fully in-browser via TensorFlow.js. Classifies with_mask, without_mask, and mask_worn_incorrect.
Open Mask DemoTeachable Machine — Mask Detector
The same 3-class face-mask demo running the locally-hosted Teachable Machine model. Test with webcam or uploaded images — no external service required.
Open TM Mask Demo ↗Vessel Classifier Demo Viewer
Open the port-operations browser demo to review sample vessel classifications, probability output, and the final transfer-learning recommendation that outperformed the baseline CNN.
Open Vessel DemoTensorFlow Playground
Google's browser-based neural network sandbox. Adjust layers, activations, learning rate, and dataset interactively — watch the decision boundary form in real time. Embedded full-size below.
Open TF Playground ↗TensorFlow Playground — embedded below. Use the controls inside the frame to experiment.
What This Portfolio Does
Explain
Each page explains objective, source data, and requirement coverage in grading-first structure.
Run
Notebooks and live demos let visitors inspect the actual model outputs, not just screenshots.
Visualize
Results render as charts, metrics, downloadable artifacts, and linked capstone materials.
Quick Launch
Use these entry points to explore the incremental capstone portfolio.