Lab Guide — RAG Basics

Load, embed, and index 10 documents in ChromaDB

The demo/sample_docs/ directory contains 10 text files about Python standard library modules. Your task:

1. Load all 10 .txt files into a list of strings
2. Embed them using sentence-transformers model all-MiniLM-L6-v2
3. Create a ChromaDB collection and add the documents with their embeddings
4. Add metadata: topic="python" for docs 1-5, topic="stdlib" for docs 6-10

Test your implementation

# Verify: query the collection
results = collection.query(query_texts=["file path handling"], n_results=3)
print(results["documents"][0])
# Should return pathlib doc as top result

Checkpoint: Collection contains 10 documents. Query returns relevant results.

Implement rag_answer(question, collection, top_k=3)

Build a function that:

1. Embeds the question using the same model
2. Retrieves the top-k most similar documents from the collection
3. Builds a prompt with the retrieved context
4. Sends the prompt to Claude and returns the answer

Prompt template

prompt = f"""Use the following context to answer the question.
If the context does not contain enough information, say so.

Context:
{context}

Question: {question}"""

Test your implementation

answer = rag_answer("How do I read a JSON file in Python?", collection)
print(answer)
# Should reference the json module documentation from the retrieved context

answer2 = rag_answer("What is a defaultdict?", collection)
print(answer2)
# Should reference the collections module

Checkpoint: Function returns grounded answers that reference specific information from the retrieved documents.

Query with metadata filters and compare results

Use ChromaDB's where parameter to filter by topic metadata.

1. Query for "iteration patterns" with where={"topic": "python"}
2. Query for the same question without any filter
3. Compare the retrieved documents — does filtering improve relevance?

Test your implementation

# Filtered query
filtered = collection.query(
    query_texts=["iteration patterns"],
    n_results=3,
    where={"topic": "python"}
)
print("Filtered:", [doc[:50] for doc in filtered["documents"][0]])

# Unfiltered query
unfiltered = collection.query(
    query_texts=["iteration patterns"],
    n_results=3
)
print("Unfiltered:", [doc[:50] for doc in unfiltered["documents"][0]])

Checkpoint: You can observe and document the difference between filtered and unfiltered results.