The 16-day AI Engineering Journey
Four phases, sixteen live sessions, four hours a day. It starts at “how does an LLM actually work” and ends with you whiteboarding an enterprise AI system — the same stack of skills behind every RAG and agent system on this site.
Start here · Foundations chapters
How software actually works, in plain words
Frontend, backend, database, dev vs production, bugs and hotfixes — the full map, explained like a mentor would, with zero heavy words. Read this first.
Building the frontend the way it works now
Claude Code setup, CLAUDE.md, feeding docs to the AI, MCPs (giving it eyes), the AI-friendly folder structure, and the plan → build → verify loop you will run every day.
Building the part nobody sees
Routes, validation, business rules, auth and status codes in plain words — with the API contract as the spec and tests as the backend's eyes.
The memory that must never lie
Tables and relationships, schema from a data-model doc, migrations, indexes, injection, backups — and why production data stays sacred.
What an LLM actually is, and how one gets built
Next-word prediction, tokens, parameters, attention, and the full build pipeline — data, pretraining, fine-tuning, RLHF — and the people behind every stage.
The knobs, the wire, and the bill
The API call, system vs user prompts, temperature, top-p and top-k, max tokens, streaming, token cost, hosted vs local — and why the key never touches the frontend.
Getting the answer you actually wanted
System prompts, zero/few-shot and chain-of-thought, forcing strict JSON, and the failure side — reducing hallucination and defending against prompt injection.
Turning meaning into numbers
Embeddings, vector space without heavy maths, and cosine similarity — how a computer measures that two different sentences mean the same thing.
Vector search, at scale
Chunking, why normal databases can't search by meaning, vector databases and nearest-neighbour search, and building real semantic search.
Answers grounded in your documents
Retrieval-augmented generation end to end: the grounding contract, citations, honest refusal, and the advanced techniques that make it reliable.
AI agents and tool calling
Tool calling, the think-act-observe loop, connecting a model to real functions, multi-agent systems, and the safety that agency demands.
Making it safe, and making it yours
Guardrails, PII redaction, rate limiting and cost control, the full capstone architecture, and the career roadmap from here to a job.
When basic RAG isn't good enough
Query expansion, hybrid search, re-ranking, better chunking and metadata — each a targeted fix for a specific retrieval failure.
Testing a thing that never answers the same way twice
Golden datasets, grading retrieval vs answer quality, LLM-as-judge, and catching regressions before your users do.
When the model is slow and the users are many
The timeout problem, streaming, async architectures and queues, WebSockets vs polling, caching, and keeping cost sane at scale.
Fine-tuning, and where you go from here
Prompt vs RAG vs fine-tuning and when to use each, how fine-tuning really works, who trains models, and the roadmap from here to a career.
Reserve a seat on the next batch
Tell me your course and where you are with AI, and I'll reply with the batch schedule, fee, and whether this journey fits where you're trying to get to.