AI Engineering 101, Volume 2: Building Production AI Systems
Turn a laptop RAG into a production service that runs 24x7. Distributed-systems essentials, the model as a system component, RAG done properly (from BM25 to Gr…
Category: Technology · Level: intermediate · 104 lessons · Price: USD 99.00
Topics: ai-engineering, system-design, rag, mcp, agents, llm
Curriculum
Module 1: Distributed Systems Essentials
- Where This Came From: From Lamport's Clocks to Raft and Jepsen
- Replication: Leaders, Followers and Quorums
- Consistency, CAP and PACELC: What a Read Can Promise
- Consensus in Practice: Raft, Failover and GitHub's 43 Seconds
- Retries Without Regret: Idempotency Keys, Jitter, Budgets and the Outbox
- Hands-on Lab: Break a Raft Cluster, Then Stop the Double Charges
- Design Exercise: Surviving a Data-Centre Failure (with Model Answer)
- Portfolio Step 6: Idempotent Ingestion and Polite Retries
- Interview Corner: Six Distributed-Systems Prompts
- Module 1 Reading List
- Module 1 Quiz: Distributed Systems Essentials
Module 2: The Model as a System Component
- Where This Came From: From Shannon's N-grams to Reasoning Models
- Tokens and Context Windows: What the Model Actually Reads
- Latency Anatomy: Loading, Prefill, Decode and Why Streaming Matters
- Cost per Request and Model Routing: The Smallest Model That Does the Job
- Rate Limits, Fallbacks and a Routing Bug: Anthropic's 2025 Postmortem
- API or Self-Hosted? Choosing Where Your Model Runs
- Hands-on Lab: Build a Model Gateway: Measure, Route, Fall Back, Budget
- Design Exercise: The Model Layer for 120,000 Students (with Model Answer)
- Portfolio Step 7: A Router, a Budget and a Usage Ledger Around app/llm.py
- Interview Corner: Six Model-Layer Prompts
- Module 2 Reading List
- Module 2 Quiz: The Model as a System Component
Module 3: GPUs and Inference Hardware
- Where This Came From: From Video-Game Graphics to the Chips Behind AI
- Will It Fit? VRAM, Quantisation and the KV Cache
- How Fast Will It Write? Prefill, Decode and Memory Bandwidth
- Serving Many People at Once: Batching, Continuous Batching and PagedAttention
- Own, Cloud or Peer-to-Peer: Choosing, Paying For and Trusting a GPU
- Hands-on Lab: Measure, Compare and Cost a GPU, Without Spending a Penny
- Design Exercise: GPU or API for 600 Students in Exam Week (with Model Answer)
- Portfolio Step 8: Point Your Study Assistant at a Rented GPU
- Interview Corner: Six GPU and Inference Prompts
- Module 3 Reading List
- Module 3 Quiz: GPUs and Inference Hardware
Module 4: RAG Done Properly, from BM25 to GraphRAG
- Where This Came From: How Search Learned to Measure, Rerank and Rewrite
- Chunking and Indexes, Properly: How You Cut Your Notes Decides What You Can Find
- Two-Stage Retrieval: Hybrid Search, Then a Reranker
- When the Question Is the Problem: Query Rewriting, Multi-Query and HyDE
- Measuring Retrieval Honestly: Recall@k, MRR and nDCG by Hand and in Excel
- Grounded Answers, GraphRAG, and When Not to Use RAG
- Hands-on Lab: Chunk, Fuse, Rerank, Rewrite and Measure on Your Laptop
- Lab Resources: The 30 Notes and 30 Labelled Questions
- Design Exercise: A Course-Notes Assistant for 12,000 Students (with Model Answer)
- Portfolio Step 9: Rerank, Refuse, Rewrite and Measure Your Study Assistant
- Interview Corner: Six Retrieval and RAG Prompts
- Module 4 Reading List
- Module 4 Quiz: RAG Done Properly
Module 5: Real-Time AI with OddSockets
- Where This Came From: From Polling and Comet to Token Streaming
- Streaming Tokens to a Browser: SSE, WebSockets and When to Use Each
- Pub/Sub and Fan-Out: Who Hears the News, and What If They Missed It?
- Connections Break: Reconnects, Heartbeats, Presence and Backpressure
- Live Progress for Long Jobs, Channel Security and the Managed Route (OddSockets)
- Hands-on Lab: Polling vs SSE vs WebSockets, Missed Messages and Backpressure
- Design Exercise: Live Features for 2,000 Students in Exam Season (with Model Answer)
- Portfolio Step 10: Live Answers and Live Progress for Your Study Assistant
- Interview Corner: Six Real-Time Prompts
- Module 5 Reading List
- Module 5 Quiz: Real-Time AI with OddSockets
Module 6: Evaluation and Observability
- Where This Came From: From BLEU Scores to LLM Judges and OpenTelemetry
- Golden Sets and Regression Evals: Testing Software That Answers in Sentences
- Measuring RAG Answers: Retrieval, Faithfulness, Citations and Refusals
- LLM-as-Judge: Biases, Controls and Checking the Judge with Cohen's Kappa
- Is the Difference Real? Confidence Intervals, Sample Sizes and Paired Comparisons
- Seeing Inside the Running System: Structured Logs, Metrics, Traces and OpenTelemetry
- Hands-on Lab: Build an Eval Suite, Catch a Regression, Check the Judge, Trace a Request
- Design Exercise: Evals and Monitoring for 120,000 Students (with Model Answer)
- Portfolio Step 11: An Eval Suite and Tracing for Your Study Assistant
- Interview Corner: Six Evaluation and Observability Prompts
- Module 6 Reading List
- Module 6 Quiz: Evaluation and Observability
Module 7: Reliability, Performance and Cost
- Where This Came From: From MTBF and Nines to Circuit Breakers and Prompt Caching
- Reliability Maths for an AI App: Repair Time, Dependencies and Fallbacks
- Timeouts, Circuit Breakers and Bulkheads Around the Model
- Caching Model Replies: Exact, Semantic and Provider Prompt Caching
- Load Testing, Percentiles and the Cost of Every Answer
- Hands-on Lab: A Hung Model, a Circuit Breaker and a Semantic Cache That Gets It Wrong
- Design Exercise: Exam Eve, a Model Outage and a $50 Budget (with Model Answer)
- Portfolio Step 12: Timeouts, a Circuit Breaker, a Reply Cache and a Cost Report
- Interview Corner: Six Reliability and Cost Prompts
- Module 7 Reading List
- Module 7 Quiz: Reliability, Performance and Cost
Module 8: Fine-Tune, RAG or Prompt?
- Where This Came From: From Transfer Learning to LoRA, QLoRA and DPO
- Fine-Tune, RAG or Prompt? A Decision Framework with Real Costs
- Datasets You Can Trust: Collection, Labelling, Dedup, Splits, Leakage and Privacy
- LoRA and QLoRA by Counting: Ranks, Parameters and Memory (with Excel)
- Did Fine-Tuning Help? Baselines, Confidence Intervals and What Breaks
- Hands-on Lab: Fine-Tune a Tiny Model on Your Laptop and Beat (or Match) the Prompt
- Design Exercise: Offline Triage for 120,000 Students (with Model Answer)
- Portfolio Step 13: A Trusted Dataset, a Fine-Tuned Triage Switch and a Decision Record
- Interview Corner: Six Fine-Tuning Prompts
- Module 8 Reading List
- Module 8 Quiz: Fine-Tune, RAG or Prompt?
Module 9: Ship It 24x7, Containers, Deployment, CI/CD
- Where This Came From: From Bare Metal to Containers, CI and GitOps
- Containers for the Study Assistant: Dockerfile, Layers, Cache and Small Images
- One Command for the Whole System: Compose, Liveness and Readiness
- Config and Secrets: Nothing Secret Inside the Image
- CI on Every Push, and Deploys You Can Undo
- Running It 24x7: Monitoring, Alerts, Runbooks and Blameless Postmortems
- Hands-on Lab: Containerise, Break and Roll Back a Small App
- Design Exercise: Exam Season, 2,000 Students, One Engineer (with Model Answer)
- Portfolio Step 14: Ship Your Study Assistant: Containers, CI, Deploy and Runbook
- Interview Corner: Six Shipping and Operations Prompts
- Module 9 Reading List
- Module 9 Quiz: Ship It 24x7
Enrol in AI Engineering 101, Volume 2: Building Production AI Systems on ClassaaS