Week 1 Mini Project - RAG Foundations Lab

Pre-reading: Foundations Overview · 01 RAG Debugging and Quality

This mini project gives Week 1 a runnable baseline. It uses only the Python standard library so you can focus on the pipeline shape: source data, chunking, retrieval, grounded answers, token budgeting, and long-context summarization.

What You Will Build

Capability	Output
Prompt builder	Context-aware prompt text
Chunker	Document slices with IDs
Retriever	Ranked chunks for a query
Grounded answerer	Answer plus citations
Long-doc summarizer	Map-reduce style summary

How to Run

cd docs/03-mini-projects/code/week01-rag-foundations

# Ask a question
python3 cli.py query "How do I reset an expired API key?"

# Summarize long text from stdin
echo "Long document text here." | python3 cli.py summarize

# Run tests
python3 -m pytest tests/ -v

Portfolio Structure

code/week01-rag-foundations/
├── chunker.py
├── retriever.py
├── grounding.py
├── summarizer.py
├── cli.py
├── corpus.json
└── tests/test_rag.py

What to Modify Across the Week

Day	Suggested change
Day 1	Change the system prompt and inspect token estimates.
Day 2	Add stronger refusal and output constraints.
Day 3	Change chunk size and compare retrieval ranking.
Day 4	Change citation formatting and context ordering.
Day 5	Log one failure and classify the root cause.
Day 6	Replace the long document text and compare summaries.
Day 7	Create a one-page debugging checklist from your notes.

Starter Assets

Asset	Purpose
week01-rag-foundations/cli.py	CLI entry point for query and summarize flows
week01-rag-foundations/chunker.py	Chunking and token estimation utilities
week01-rag-foundations/retriever.py	Ranking and top-k retrieval logic
week01-rag-foundations/grounding.py	Grounded prompt and citation answer pipeline
week01-rag-foundations/summarizer.py	Long-context map-reduce summarizer
week01-rag-foundations/corpus.json	Tiny document set for retrieval
week01-rag-foundations/tests/test_rag.py	Unit test suite for RAG building blocks
week01_rag_foundations_output.txt	Sample retrieval and long-context output from a successful run

Matching Lab Outputs

Output	Why keep it
Retrieval ranking snapshot	Helps explain why an answer failed or succeeded
Token estimate summary	Connects design decisions to cost and latency
Failure record	Becomes a regression test seed
Long-doc summary	Shows handling of context compression

Portfolio Checklist

Item	Done?
Save one retrieval result with chunk IDs and citations.	[ ]
Save one failed query case and root-cause notes.	[ ]
Capture one chunk-size comparison and trade-off summary.	[ ]
Write one STAR-ready bullet using this project evidence.	[ ]