Ethics, Security & Capstone

Module 08 · Days 50–60

Responsible AI is a market requirement. Bias, safety, security, privacy — then your Training Phase capstone: a portfolio-grade full-stack AI product.

D50

AI bias, fairness metrics & model cards

Measuring and mitigating algorithmic bias. Fairness metrics, protected attributes, and model cards.

D51

AI security: prompt injection, jailbreaks & red-teaming

How AI systems are attacked: prompt injection, jailbreaks, data poisoning — and practical defences.

D52

Privacy: GDPR for ML engineers & PII handling

What GDPR means for ML. PII detection, right-to-erasure in vector stores, differential privacy basics.

D53

AI infrastructure cost optimisation

Quantisation, speculative decoding, batching, vLLM — making AI 5 to 10 times cheaper without losing quality.

D54

Portfolio strategy & technical storytelling

How hiring managers read ML portfolios. READMEs that impress, LinkedIn positioning, demo videos.

D55

Capstone planning: scope, architecture & success metrics

Design your final capstone. Problem statement, tech stack, evaluation approach, architecture diagram.

D56

Capstone build — Day 1: data pipeline & model baseline

Data ingestion, preprocessing, model selection, and a working evaluation baseline.

D57

Capstone build — Day 2: API, UI & test suite

FastAPI backend, Gradio or Streamlit frontend, and full test suite passing in CI.

D58

Capstone build — Day 3: deployment & monitoring

Docker, cloud deployment, Prometheus metrics, and Evidently monitoring report.

D59

Capstone build — Day 4: documentation & demo prep

README, model card, architecture diagram, and a scripted 2-minute demo walkthrough.

🏗

🏗️ FINAL PROJECT: Full-Stack AI System Capstone

A portfolio-grade AI product combining 3 or more skills from the Training Phase. Deployed, evaluated, documented.