LangChain in 10 Days

A beginner bootcamp that runs entirely on free AI APIs.

No paid account and no GPU. If you can write a for loop, you can do this. Days 9 and 10 use Docker to run the vector database, which is one command.

This course lives inside a larger training repo. Every command below runs from this folder, 2026/langchain-10-days, not from the repo root.

---

The ten days

Each day is one slide deck and one notebook, about 60 to 90 minutes.

Day	Title	What it covers	You end up with
1	Your First AI Program	ChatOpenRouter, invoke, the reply object, tokens and cost, temperature, max_tokens	A script that answers a question
2	Roles: System, Human, Assistant	The model sees a list of messages, what a long system message costs, pretty_print	The same assistant with three personalities
3	Prompt Templates	ChatPromptTemplate, placeholders, the KeyError, literal braces, partial, MessagesPlaceholder	A reusable email-reply writer
4	Chains: The Pipe	The pipe operator, StrOutputParser, invoke and batch and stream, fallbacks	A summariser chain
5	Structured Output	Pydantic models, with_structured_output, Literal, why optional fields come back empty	A review reader that returns an object
6	Memory	Checkpointers, thread_id, tokens growing every turn, trim_messages, summarisation, SqliteSaver	A chatbot that remembers
7	Tools	@tool, the schema the model sees, bind_tools, ToolMessage, a live web search	An assistant with a calculator and web search
8	Your First Agent	create_agent, reading the trace, what an agent costs, ModelCallLimitMiddleware	An agent that plans two steps
9	Documents, Chunks and Vectors	Splitting, embeddings, similarity by hand, Qdrant, metadata filters, where vector search fails	A search that works by meaning
10	RAG with RAGWire	One config file, ingest, retrieve, citations, refusal, top_k, an agent that decides when to search	A RAG assistant over two real 10-K filings that cites its pages

Days 1 to 8 need nothing installed beyond Python. Day 9 downloads a small embedding model, about 130 MB, and runs it on your CPU. Days 9 and 10 need Docker for the vector database.

---

Setup, once

1. Get a free API key

Go to openrouter.ai/keys, sign in, and create a key. It starts with sk-or-.

OpenRouter gives you access to many models through one key. The models this course uses end in :free, so you are not charged.

Free models have a daily limit. If a cell suddenly starts failing with a 429, you have hit it. Wait, or switch to another :free model. It is not your code.

From Day 7 you also need a second free key, for the web search tool: ollama.com/settings/keys. Days 1 to 6 do not use it, so you can leave it until then.

2. Install Python packages

This project uses uv. If you do not have it:

winget install astral-sh.uv

On macOS or Linux:

curl -LsSf https://astral.sh/uv/install.sh | sh

Then, from the project folder, one command does everything:

uv sync

That creates .venv, installs the right Python if you do not have it, and installs every package at the exact version recorded in uv.lock. You do not need to activate anything.

Run things with uv run, which uses that environment without you activating anything:

uv run jupyter notebook notebooks

3. Save your keys

Copy .env.example to a new file called .env, and paste your keys in:

OPENROUTER_API_KEY=sk-or-...your key...
OLLAMA_API_KEY=...your key...

.env is listed in .gitignore, so it will never be committed. Never paste a key into a notebook cell, and never print one.

4. Start the vector database (needed from Day 9)

Days 1 to 8 need nothing here. Days 9 and 10 store vectors in Qdrant, which runs in Docker. Install Docker Desktop, then from the project folder:

docker compose up -d

Check it came up:

docker compose ps

You can also open http://localhost:6333/dashboard in a browser and see the collections as you create them. To stop it later, docker compose down. Your data stays in a Docker volume, so starting it again keeps everything.

5. Register the kernel, once

So that Jupyter runs this environment and not some other Python on your machine:

uv run python -m ipykernel install --user --name lc10 --display-name "Python (langchain-10-days)"

6. Start Jupyter

uv run jupyter notebook notebooks

Open Day01_First_AI_Program.ipynb, choose Python (langchain-10-days) from the Kernel menu, and run the first cell.

---

How to use each day

1. Open slides_pdf/DayNN_Slides.pdf. Read the pictures.
2. Open notebooks/DayNN_*.ipynb and run the cells one at a time.

The slides hold the diagrams and the explanation. The notebooks hold the code. Neither repeats the other.

slides_pdf/ also has three longer talks that go past the ten days: Graph Engineering, Loop Engineering and Memory Engineering. They are background reading, not part of any day.

---

What is in this folder

langchain-10-days/
├── README.md               this file
├── pyproject.toml          every package, pinned
├── .env.example            copy to .env, add your free keys
├── docker-compose.yml      Qdrant, for Days 9 and 10
├── notebooks/
│   ├── Day01_*.ipynb ... Day10_*.ipynb
│   ├── config/             the Day 10 RAG pipeline, in YAML
│   ├── data/               three short documents, for Day 9
│   └── filings/            two real 10-K PDFs, for Day 10
└── slides_pdf/             one deck per day

---

If something breaks

What you see	What it means
OPENROUTER_API_KEY not found	Your .env is missing, or it is not in the notebooks folder
401	The key is wrong. Copy it again, with no spaces
429	You hit the free per-minute limit. Wait a minute and run the cell again
ModuleNotFoundError	Jupyter is on the wrong kernel. Pick Python (langchain-10-days), or run uv sync again
Day 7 web search fails to authenticate	OLLAMA_API_KEY is missing from .env. It is a separate free key
Day 9 is slow the first time	It is downloading a small embedding model. Once only
Every HTTPS call hangs until it times out	You are behind a TLS-inspecting proxy. uv sync installs truststore, which fixes it
Connection refused on port 6333	Qdrant is not running. docker compose up -d from the project folder
docker: command not found	Docker Desktop is not installed, or not started

---

KGP Talkie · Laxmi Kant Tiwari YouTube · LinkedIn · kgptalkie.com