AI Engineering: Build AI-Powered Apps

Your first LLM API call (free Gemini API)

Get a free Gemini API key, make your first call from Python, control the answer with a config, build a small memory-enabled chatbot and handle rate-limit errors. This one call is the base of every AI feature you will build.

Tokens, cost, latency and choosing a model

Tokens decide what an LLM call costs, how long it takes and how much fits in one request. Learn to count them, estimate cost, cut latency and pick the cheapest model that does the job.

Streaming responses to a UI

Stream a model's answer token by token: first in the terminal, then through a FastAPI endpoint to a plain JavaScript page with a Stop button. Learn why it feels faster and how to handle errors mid-stream.

Structured output with JSON schemas

Make the model return data your code can trust. Define a schema with Pydantic, request JSON output from the Gemini API, validate the values and handle missing information, using a coffee-order parser as the example.

Function calling: letting the model use tools

Let a model look things up and take actions by calling your Python functions. Build a Brew & Bean assistant with automatic and manual function calling, and learn how to keep tool use safe.

Embeddings and semantic search

Embeddings turn text into vectors so a computer can compare meaning, not just words. Learn what they are, how cosine similarity works and build a small semantic search over menu items with the Gemini API.

Vector databases: pgvector, Chroma and friends

Store embeddings in a vector database so search is persistent, filterable and fast. Build the same menu search with Chroma and with PostgreSQL plus pgvector, and learn how to choose between them.

Retrieval-augmented generation (RAG) end to end

Build a small RAG assistant in Python: chunk your own files, embed them, retrieve the best passages and make the model answer only from them, with citations and an honest 'I don't know'.

Chunking, hybrid search and re-ranking

Most RAG failures are retrieval failures. Learn how to chunk documents, combine keyword and vector search with Reciprocal Rank Fusion, filter with metadata, and re-rank so the right passage reaches the model.

Project: coffee shop menu assistant with RAG

Build a Brew & Bean menu assistant: hybrid search over menu items and FAQs, a grounded answerer that cites sources, a price check guardrail, a FastAPI endpoint and a retrieval hit-rate test.

AI agents: tools, loops and planning

An agent is an LLM in a loop that chooses tools, reads the results and decides what to do next. Build one in about 50 lines of Python, then learn how agents fail and how to keep them safe and affordable.

Model Context Protocol (MCP): connecting AI to tools

MCP is an open standard that lets any AI app use your tools and data. Build a Brew & Bean MCP server in Python with tools, a resource and a prompt, test it with a client, and connect it to an AI app safely.

Project: build a research agent

Build a research agent in about 100 lines of Python with the free Gemini API and Wikipedia: it plans, searches, reads, saves notes, and writes a sourced report, with step limits, error handling and checks against invention.

Multi-agent systems and when to avoid them

Multiple agents can split work, but they cost more, lose context and fail in harder-to-debug ways. Learn the main patterns, build a drafter and reviewer, and use a decision ladder to choose the simplest design that works.

Evaluating AI apps (evals)

Build a small Python eval harness: a test set, code checks, an LLM judge and a score you can compare after every prompt or model change, so you catch regressions before your users do.

Guardrails, safety and prompt-injection defence

Models can be tricked, and they also make honest mistakes. Learn the layered defences an AI engineer builds: input checks, least-privilege tools, output validation and tests, so a fooled model cannot do real damage.

Running open models locally with Ollama

Install Ollama, run a small open model on your own laptop, call it from Python (chat, streaming, structured output and OpenAI-style code), and learn when local models are a better choice than a cloud API.

Prompting vs RAG vs fine-tuning

Your model gives wrong or off-style answers. Should you fix the prompt, add retrieval or fine-tune? Learn what each approach changes, a simple decision order, and why fine-tuning is the last step for most apps.

Production: caching, cost control and monitoring

Turn an AI demo into something you can run for real users: cache answers, cap tokens and budgets, add retries and fallbacks, and monitor cost, latency and quality without leaking personal data.