AI engineer with Python
Build LLM apps, RAG, agents and MCP servers in Python, then measure them.
At the end: Ship an AI feature end to end: prompts, retrieval, tools, evals and the harness around the model.
Stage 1: Foundations
The language and the model, before any framework.
- Python, properly: Types, async and tests: the Python every AI lab after this uses.
- Building with LLMs: Messages, tokens, streaming, tools and structured output, on the raw API.
Stage 2: Core AI engineering
The four things every AI product is made of.
- RAG in Python: Search that answers from your own documents, with citations.
- Agents from scratch: The agent loop from scratch: tools, budgets, planning and permissions.
- MCP servers and clients: Give any assistant your tools through the Model Context Protocol.
- Evals that catch regressions: Catch a regression before your users do.
- Memory and multi-agent systems: Remember across sessions and split work between agents.
- Harness engineering: Everything around the model that makes an agent reliable.
- AI security: attack and defend: Break your own agent, then defend it, before an attacker does.
- Build your own coding agent: Build the coding agent itself: tools, safe edits, a sandbox and a loop that fixes.
- Claude and the Anthropic SDK: Claude the right way: tools, caching, thinking and batches.
- Build an AI gateway: Run AI in production: one gateway over many providers, within budget and SLO.
- AI engineering traps: The AI bugs interviews ask about, found and fixed.
- Fine-tuning basics: When to fine-tune, and how: LoRA on a model you can see inside.
- Build a voice agent: A phone agent that listens, answers in time and knows when to stop talking.
Stage 3: Frameworks
Now that you know what they do for you.
- LangChain and LangGraph: Chains and graphs with LangChain and LangGraph.
- Pydantic AI: Typed agents with Pydantic AI.
- OpenAI Agents SDK: Handoffs, guardrails and tracing with the OpenAI Agents SDK.
Stage 4: Build it for real
Whole AI apps for a company, one feature at a time.
- AI support copilot project: A support copilot that answers from policy and knows when to hand over.
- AI ordering agent project: An ordering agent that acts, safely.
- AI invoice reader project: Read invoices into checked, structured data.
- MCP server project: A production MCP server.
- AI answer grader project: An AI grader you can trust, with evals.
- Build your own GPT project: Build a small GPT yourself and see how the model works inside.
Stage 5: Interview ready
The bugs interviewers ask about.
- Python traps: Reproduce, fix and explain real Python bugs.