← Back to Resources

By the TechStudio editorial team · Updated September 25, 2026 · Editorial policy

8-Week Roadmap AI Engineering Free Learning Resources

8-Week AI Engineer Roadmap

The engineers building production-ready AI systems are not just learning more. They are learning what to build, how the pieces fit together, how to evaluate them, and how to ship them. This roadmap turns that idea into an eight-week sequence of concepts, implementation work, free resources, portfolio deliverables, and interview preparation.

The loop
Learn→ Build→ Deploy→ Prepare→ Apply

Roadmap at a glance

Eight weeks, eight engineering milestones

01
LLMs
Mental model + internals
02
RAG
Build from scratch
03
APIs + Vectors
Embeddings + services
04
Agents
LangGraph + CrewAI
05
Claude Code + MCP
Agentic workflows
06
LLMOps
Evals + monitoring
07
Deploy
Docker + FastAPI
08
Career
Portfolio + interviews

Week-by-week plan

What to learn, build and finish each week

01
Week 1

LLMs & how they actually work

Build a correct mental model before jumping into frameworks. Understand the path from text to tokens, embeddings, transformer computation, logits and generated tokens.

Learn

  • Tokens, tokenization and context windows
  • Embeddings and positional information
  • Attention and transformer blocks
  • Pretraining, instruction tuning and alignment
  • Inference, temperature, top-p and decoding
  • Hallucination, tool use and limitations

Build

  • Tokenization explorer in Python
  • Simple next-token prediction demo
  • Prompt playground with controlled parameters
  • Small CLI chat application

Finish with

  • One-page LLM architecture notes
  • Five common LLM failure modes
  • GitHub README with diagrams
  • Short explanation you can give in an interview
Free video
Andrej Karpathy — Intro to Large Language Models ↗

A one-hour technical overview covering inference, training, fine-tuning, tool use, multimodality and LLM security.

02
Week 2

Build your first RAG pipeline from scratch

Learn the complete retrieval-augmented generation flow instead of treating RAG as a single library call.

Learn

  • Document loading and cleaning
  • Chunking strategies and metadata
  • Embedding models and vector similarity
  • Retrieval, context assembly and prompting
  • Grounding, citations and failure modes
  • Basic retrieval and answer evaluation

Build

  • PDF or Markdown knowledge base
  • Embedding + vector store pipeline
  • Retriever + generator
  • Source-aware answer format
  • Five-question evaluation set

Go deeper

  • Hybrid search
  • Reranking
  • Query rewriting
  • Parent-child retrieval
  • Context compression
Free video + code
LangChain — RAG From Scratch video playlist ↗•GitHub notebooks ↗

The playlist and companion repository walk through indexing, retrieval, generation and advanced query transformations.

03
Week 3

APIs, vector databases & embedding models

Move from notebook experiments to services. Learn how an AI application talks to models, stores vectors and exposes functionality through APIs.

Learn

  • REST APIs, JSON and HTTP
  • Authentication and environment variables
  • Embeddings: dimensions and similarity
  • Vector indexes and metadata filters
  • Latency, throughput and cost
  • Async vs synchronous API calls

Build

  • FastAPI backend for your RAG app
  • /health and /query endpoints
  • Vector search service
  • Typed request/response models
  • Error handling and logging
YouTube practice
Free YouTube search: embeddings + vector databases + RAG ↗

Use this as a topic index and prefer recent, code-along material when APIs or SDKs have changed.

04
Week 4

AI Agents & orchestration with LangGraph + CrewAI

Understand the difference between a fixed pipeline and a system that can choose tools, maintain state, route work, retry and coordinate multiple steps.

Learn

  • Tool calling and structured outputs
  • State, memory and checkpoints
  • Nodes, edges and conditional routing
  • Loops, retries and stopping conditions
  • Single-agent vs multi-agent workflows
  • Human-in-the-loop patterns

Build

  • Research agent with web/search tool
  • RAG agent with retrieval tool
  • LangGraph state machine
  • Small CrewAI multi-agent workflow
  • Retry + validation node
Free videos
LangGraph Explained for Beginners ↗Free YouTube search: CrewAI multi-agent tutorial ↗
05
Week 5

Claude Code, MCP & agentic workflows

Learn how coding agents and MCP-style tool connections fit into an AI engineering workflow. The goal is not just using an agent, but understanding permissions, tools, context, state and repeatable workflows.

Learn

  • Agentic coding workflows
  • Plan → implement → test → review loops
  • Claude Code basics and project instructions
  • MCP clients, servers, tools and resources
  • Permissions and safe tool execution
  • Context engineering for agents

Build

  • Use Claude Code on a small repo
  • Create an MCP server with one useful tool
  • Connect an agent to a local resource
  • Automate one repeatable development task
Free videos
Tech With Tim — Claude Code Full Tutorial for Beginners ↗Anthropic — The Model Context Protocol (MCP) ↗
06
Week 6

LLMOps, evaluations & production monitoring

Treat an AI application like a system that must be tested and observed. Learn how to detect regressions instead of relying on subjective “it looks good” checks.

Learn

  • Golden datasets and test cases
  • Correctness, relevance and groundedness
  • Rule-based and model-graded evaluation
  • Trace collection and debugging
  • Latency, cost and error monitoring
  • Offline vs online evaluation

Build

  • 20–50 question evaluation set
  • Automated regression checks
  • Trace every RAG/agent step
  • Latency and token/cost logging
  • Failure dashboard or report
Free video
Hamel Husain — Instrumenting & Evaluating LLMs ↗

A detailed lesson covering unit tests, LLM-as-judge, human evaluation, traces, LangSmith, datasets and agent evaluation.

07
Week 7

Docker + FastAPI: containerize & deploy your AI stack

Turn the project into a service another person can run. Learn containerization, API contracts, configuration, health checks, logs and a basic cloud deployment flow.

Learn

  • Dockerfile, images and containers
  • Ports, volumes and environment variables
  • FastAPI routes and Pydantic models
  • Health checks and structured errors
  • Logs, secrets and basic security
  • Cloud deployment concepts

Ship

  • Dockerize your RAG or agent app
  • Expose /health and /query
  • Add a production-style README
  • Test locally with Docker
  • Deploy to a cloud platform
Free videos
Deploy FastAPI with Docker — beginner guide ↗From Zero to Production: Docker + FastAPI + LLMs ↗
08
Week 8

Portfolio + interview preparation + job applications

Convert the previous seven weeks into evidence. Your goal is not a huge collection of unfinished demos; it is a small set of projects you can explain deeply.

Portfolio

  • 2–3 polished AI projects
  • Architecture diagram for each
  • README with setup and screenshots
  • Evaluation results and limitations
  • Deployment URL when possible
  • Clean GitHub history and structure

Interview

  • Explain one RAG system end to end
  • Explain an agent state graph
  • Discuss retrieval and eval metrics
  • Discuss latency, cost and scaling
  • Debug hallucinations systematically
  • Practice AI system design

Applications

  • Tailor resume to each role
  • Lead with measurable project outcomes
  • Link GitHub and deployed demos
  • Prepare a 60-second introduction
  • Track applications and follow-ups
  • Iterate from interview feedback
Free career resources
GitHub portfolio basics ↗Free YouTube: AI engineer interviews ↗Free YouTube: LLM/RAG system design ↗Free YouTube: AI engineer resume/projects ↗

Project ladder

Build these alongside the roadmap

Project 1

Production-style RAG assistant

PDF ingestion → chunking → embeddings → vector search → reranking → grounded answer → citations → evaluation → FastAPI → Docker.

Project 2

Agentic research assistant

Planner → search tools → source extraction → quality check → retry/reformulation → synthesis → trace/evaluation.

Project 3

MCP-powered productivity agent

Expose useful tools through MCP, connect an agent, enforce permissions, log tool calls, and add human approval for sensitive actions.

Project 4

AI evaluation dashboard

Create a dataset, run repeatable evaluations, compare prompt/model versions, inspect traces, and track latency, errors and cost.

Daily execution plan

A simple weekly rhythm

Day 1

Concepts + notes

Day 2–3

Code-along + implementation

Day 4–5

Build your own version

Day 6

Test, evaluate, document

Day 7

Review + publish + plan next week

Rule: do not spend the whole week consuming tutorials. Every week should produce something runnable, measurable or explainable.

Interview checklist

By the end of week 8, be able to explain these

LLM fundamentals

Tokens, attention, context windows, decoding, embeddings, fine-tuning, tool calling, hallucinations and prompt injection.

RAG system design

Chunking, embeddings, vector search, hybrid retrieval, reranking, metadata filters, context construction, citations and evaluation.

Agent architecture

Tools, state, memory, routing, loops, retries, human-in-the-loop, multi-agent orchestration, guardrails and observability.

Production engineering

API design, Docker, secrets, latency, concurrency, logging, monitoring, CI/CD, cost control and failure recovery.

Important: the roadmap is a sequence, not a guarantee

Eight weeks can give you a focused foundation and several portfolio artifacts, but production AI engineering requires continued practice. APIs, frameworks and model capabilities change quickly, so always check the current official documentation before implementing a production system.

Start today

Learn → Build → Deploy → Prepare → Apply

Pick Week 1, create a GitHub repository for the roadmap, and publish something small by the end of the first week. Progress compounds when every week leaves behind working evidence.

Back to Resources →