MLOps, Production & Product Building Notes
Subject Overview
The final taught semester takes models out of notebooks and into real production systems that users depend on — backend engineering for AI, deployment and infrastructure, MLOps and LLM ops, and the product thinking that separates a demo from something people pay for. It closes on responsible AI, regulation, and where the frontier is heading.
Subject at a glance
8
Units
51
Topics
15
Hours
8
Infographics
Unit-wise Syllabus
8 units — click WhatsApp below to get the full notes for each
Unit 1: Backend for AI (3 hours)
- FastAPI
- async Python
- request handling
- streaming responses
- authentication
- rate limiting
- 1
Client sends a request
- 2
Authenticate
API key or token
- 3
Rate limit
- 4
Validate input
Pydantic models
- 5
Call the model
Async
- 6
Stream the response back
Study guide
This module covers backend for AI: FastAPI, async Python, request handling, streaming responses, authentication and rate limiting. Streaming makes AI apps feel fast, so learn to implement it.
Unit 2: Deployment & Infrastructure (4 hours)
- Docker
- container registries
- deploying to Modal
- Replicate
- AWS
- GCP and Vercel
- GPU serving
- 1
Write a Dockerfile
- 2
Build the image
- 3
Push to a container registry
- 4
Deploy
Modal, AWS, GCP, Vercel
- 5
Attach GPUs if needed
- 6
Expose an endpoint
Study guide
You study deployment and infrastructure: Docker, container registries, deploying to Modal, Replicate, AWS, GCP and Vercel and GPU serving. Containerise your project so it runs the same everywhere.
Unit 3: MLOps Foundations (3 hours)
- CI/CD for ML
- model versioning
- monitoring drift
- A/B testing
- feedback loops
- cost management
- 1. Train: Versioned data and code
- 2. Validate: Automated tests in CI
- 3. Deploy: Versioned model
- 4. Monitor: Drift, latency, cost
- 5. Collect feedback: A/B tests, user signals
Study guide
MLOps foundations include CI/CD for ML, model versioning, drift monitoring, A/B testing, feedback loops and cost management. Models degrade over time, so monitoring is part of the job.
Unit 4: LLM Ops & Evaluation (3 hours)
- Prompt versioning
- eval pipelines
- observability
- guardrails
- caching
- prompt injection defenses
Prompt versioning
Track every change
Eval pipelines
Run on every pull request
Observability
Traces, tokens, latency
Guardrails
Input and output checks
Caching
Cut cost and latency
Study guide
LLM Ops covers prompt versioning, evaluation pipelines, observability, guardrails, caching and prompt injection defences. Treat prompts like code: version, test and review them.
Unit 5: AI Product Thinking (4 hours)
- Identifying use cases where AI wins
- UX patterns for AI products (streaming, citations, undo)
- pricing and unit economics
- the 'demo versus product' gap
Users
You, on the happy path
Real users with messy input
Errors
Ignored
Handled gracefully, with undo
Trust
Not needed
Citations, streaming, transparency
Economics
Free credits
Cost per user below price
Study guide
AI product thinking asks where AI genuinely wins, UX patterns for AI (streaming, citations, undo), pricing and unit economics and the gap between a demo and a product. Calculate cost per request before launch.
Unit 6: Ethics, Safety & The Frontier (3 hours)
- Responsible AI
- regulation (EU AI Act and others)
- keeping up with research
- what's next — reasoning models
- agents
- embodied AI
- Unacceptable risk
Banned, e.g. social scoring
- High risk
Strict rules: hiring, credit, medical
- Limited risk
Transparency duties, e.g. chatbots
- Minimal risk
Few obligations, e.g. spam filters
Study guide
You study ethics, safety and the frontier: responsible AI, regulation such as the EU AI Act, keeping up with research and what is next, such as reasoning models, agents and embodied AI. Learn the principles behind regulation, since the details keep changing.
Unit 7: Tools introduced
- FastAPI
- Docker
- GitHub Actions
- Modal
- Replicate and Hugging Face Inference Endpoints for model serving
- Vercel
- Railway
- Fly.io for app hosting
- Posthog
- Sentry
- Langfuse for product and LLM observability
- Stripe for monetization basics
FastAPI and Docker
Build and package services
GitHub Actions
CI/CD
Model serving
Modal, Replicate, HF Endpoints
App hosting
Vercel, Railway, Fly.io
Observability
PostHog, Sentry, Langfuse
Payments
Stripe
Study guide
The tools include FastAPI, Docker, GitHub Actions, Modal, Replicate, Hugging Face Inference Endpoints, Vercel, Railway, Fly.io, Posthog, Sentry, Langfuse and Stripe. Set up error tracking and analytics on a live project.
Unit 8: Mini-projects
- Deploy a fine-tuned model behind a FastAPI service with auth and logging
- build a complete CI/CD pipeline for an LLM app with automated evals on each PR
- launch a polished AI product on Product Hunt with payment
- monitoring and feedback loops
- 1
Deploy behind auth and logging
- 2
Add automated evals in CI
- 3
Integrate payments
- 4
Set up monitoring and feedback
- 5
Launch publicly
Product Hunt
- 6
Iterate on real usage
Study guide
The mini-projects are a fine-tuned model behind a FastAPI service with auth and logging, a CI/CD pipeline with automated evals on each PR and a polished AI product launched on Product Hunt with payments and monitoring. Launching publicly, even small, is the strongest portfolio move.
Why MLOps, Production & Product Building Matters
Anyone can build a notebook demo, but companies hire people who can deploy, monitor and improve AI in production. This module turns your models into real, maintainable products.
How to Prepare for MLOps, Production & Product Building
- Deploy something real and keep it running for at least a month.
- Add automated evaluations to your repository's pipeline.
- Track cost, latency and errors from day one.
MLOps, Production & Product Building: Frequently Asked Questions
What is MLOps?
MLOps is the set of practices for deploying, monitoring and maintaining machine learning models reliably in production.
What is model drift?
Drift happens when real-world data changes so that a model's performance gradually declines.