Fine-Tuning vs RAG for a Large Document Knowledge Base

MediumLLMs & GenAI

Reported in interviews at Mistral AI

Problem statement

How would you answer questions from a technical manual that is too large to read at once? Users need exact codes, parts and procedures with source sections, and the manual changes regularly. Compare fine-tuning with retrieval, handle exact identifiers, and explain how you would build and evaluate the system.

Show hint

Ask what fine-tuning is actually good at teaching a model, and what it is bad at.

Relevant roles: AI Engineer, Gen AI Engineer, LLM Engineer, AI Architect and Forward Deployed Engineer.

Related free problems