Fix Broken RAG Pipeline

Debuggingmedium~30 min

Objective

Fix three bugs so the RAG pipeline returns relevant, grounded answers.

Background

You've inherited a RAG pipeline that was working last week but now returns irrelevant or hallucinated answers. The team suspects recent changes to the chunking strategy and retrieval logic introduced regressions. Your job is to find and fix the bugs.

Requirements

1.Fix the chunk size to preserve semantic context (currently too large)
2.Add embedding normalization before cosine similarity search
3.Fix the prompt template variable mismatch in retrieval

Evaluation (100 points)

Chunk size is reasonable

chunk_size should be ≤ 1000 for effective semantic retrieval

20pt

Embeddings are normalized

Vectors must be L2-normalized before cosine similarity

25pt

Prompt template variables match

Template placeholders must match the format() arguments

20pt

Main uses reasonable config

The main() function should use a reasonable chunk_size

15pt

Retrieval uses proper cosine similarity

The retrieve method should compute correct cosine similarity

20pt

Hints

Select a file to start editing

Terminal

AI Assistant50K tokens left

Ask me about the code, bugs, or concepts.
I'll guide you in plain English, no code output.
Budget: 50K tokens per lab