Projects
Build. Ship. Iterate.
A portfolio-driven approach: each project follows a full pipeline from data to production, with clean structure and reproducibility.
Featured Projects
Examples of production-oriented deliverables students can build during programs.
Telco Customer Churn — MLOps API
Train a churn classifier, track experiments, package the service, and deploy a reliable prediction API with monitoring signals and validation.
Sample deliverables
- Reproducible ML project template (train/infer split)
- FastAPI endpoint + OpenAPI docs + schema validation
- Docker image + run instructions + basic monitoring plan
Computer Vision Quality Inspection
Fine-tune a CNN with transfer learning, augmentation, and robust evaluation, then export an inference pipeline for batch scoring.
Sample deliverables
- Training notebook + checkpointing + metrics tracking
- Error analysis (confusion matrix + failure clusters)
- Batch inference script + export format for production
GenAI Assistant with RAG
Build a retrieval-augmented assistant with chunking, embeddings, vector search, evaluation routines, and safe prompting patterns.
Sample deliverables
- RAG pipeline (ingest → index → retrieve → answer)
- Evaluation set + scoring rubric for reliability
- Prompt injection awareness + mitigation checklist
Time Series Forecasting (Demand & KPI)
Forecast weekly demand with proper backtesting, strong baselines, feature engineering, and a delivery-ready output format.
Sample deliverables
- Backtesting strategy + baseline comparison report
- Forecast export (CSV/API payload format)
- Monitoring signals for drift and seasonality shifts
NLP Ticket Routing with Transformers
Fine-tune a transformer classifier to automatically route support tickets, with strong error analysis and threshold tuning.
Sample deliverables
- Fine-tuned model + evaluation notebook
- Confusion-matrix driven error analysis report
- Inference pipeline template (tokenization + batching)
Fraud Detection — Cost-Aware Decisions
Build a fraud model and define a business decision policy using cost-sensitive metrics, calibration, and optimized thresholds.
Sample deliverables
- Cost-aware evaluation (precision/recall tradeoffs)
- Probability calibration + threshold policy
- Short decision memo for stakeholders
Recommendation System — Baselines & Ranking
Build recommender baselines and evaluate offline with proper splits for implicit feedback, ranking metrics, and cold start notes.
Sample deliverables
- Popularity + similarity baseline recommenders
- Offline evaluation report (ranking metrics concepts)
- Product constraints note (cold start + feedback loop)
Data Quality & Validation Suite
Create automated data validation checks and monitoring signals for pipelines feeding ML systems (parity, drift, anomalies).
Sample deliverables
- Validation rules (ranges, nulls, duplicates, schema)
- Drift signals proposal (features + thresholds)
- Incident checklist (alerts → investigation → fix)
MLflow Tracking & Model Registry
Standardize experiment tracking, compare runs, register models, and define a simple promotion workflow (staging → prod).
Sample deliverables
- MLflow project template + logging conventions
- Model registry workflow with version promotion
- Release notes template for model changes
Executive Analytics Dashboard
Transform raw data into a decision-ready dashboard with a KPI tree, drill-down pages, and a clear narrative for stakeholders.
Sample deliverables
- Dashboard pages (KPI overview + drill-downs)
- Data model documentation + key measures list
- Insights deck (5–8 slides) for decision makers
Want to see a full demo?
We can share a sample repo structure and a live walkthrough.