AI-2: AI Backend Engineering

Build AI-powered pipelines in Python. Extract structure from unstructured inputs, work with embeddings and vector databases, and implement retrieval-augmented generation (RAG) to improve model output.

First-Time Setup

1. Clone this repo using GitHub Desktop (URL provided in class)
2. Open the folder in VS Code (File → Open Folder → select odsa-ai-backend)
3. Open the setup notebook: notebooks/session_0_1_setup.ipynb
4. Follow it step by step — it will install everything and verify your environment

The setup notebook walks you through installing uv, Python, dependencies, selecting your kernel, and creating your workspace. Every step has a verification check — all should show PASS before you move on.

Before Each Class

1. Open GitHub Desktop → Fetch origin → Pull
2. Copy the new session notebook from notebooks/ into my_work/
3. Open the copy from my_work/ and work there

Why my_work/? This folder is git-ignored. When the instructor pushes updates, git pull updates the master copies in notebooks/ without touching your work. No merge conflicts, no lost progress.

Project Structure

odsa-ai-backend/
├── notebooks/           ← Master copies (don't edit these directly)
│   ├── session_0_1_setup.ipynb
│   ├── session_1_1.ipynb
│   └── ...
│
├── my_work/             ← YOUR workspace (git-ignored, safe from updates)
│   ├── session_1_1.ipynb
│   └── ...
│
├── src/                 ← Pre-built pipeline modules (provided to you)
│   ├── s0_generation/   ← Session 1.1: API integration
│   ├── s1_extraction/   ← Session 1.2: Batch processing
│   ├── s2_embeddings/   ← Session 2.1: Embeddings
│   ├── s3_ingestion/    ← Session 2.2: Vector ingestion
│   ├── s4_retrieval/    ← Sessions 3.1–3.2: RAG
│   └── s5_observability/← Session 4.1: Logging
│
├── data/                ← Course dataset
├── pyproject.toml       ← Dependencies (uv reads this)
├── .env.example         ← API key template
└── .gitignore           ← Keeps .env, .venv/, my_work/ out of git

Sessions

Week	Session	Topic
1	1.1	LLM API Integration
1	1.2	Batch Processing & Extraction
2	2.1	Embeddings & Model Selection
2	2.2	Chunking & Vector Ingestion
3	3.1	Naive RAG
3	3.2	Metadata-Aware RAG
4	4.1	Observability & Debugging
4	4.2	Module Test

Grading

Component	Weight
Lab 1 — Batch Extraction	20%
Lab 2 — RAG Evaluation	20%
Written Exam	50%
Participation	10%
Passing	70% or above

Troubleshooting

Problem	Fix
uv not found	Close and reopen your terminal after installing uv
Kernel not listed in VS Code	Cmd/Ctrl+Shift+P → "Reload Window", then reselect kernel
"Module not found" error	Make sure your kernel is set to .venv, not system Python
AuthenticationError	Check .env is in the project root and key starts with sk-ant-
Packages missing after uv sync	Reload VS Code window, then re-run the package check cell
Cells error after reloading VS Code	Normal — reload clears variables. Run cells from the top again
git pull has conflicts	You may have edited a file in notebooks/. Work in my_work/ instead

Need Help?

• Re-run notebooks/session_0_1_setup.ipynb to diagnose environment issues
• Check the troubleshooting table above
• Ask in class or during office hours