WWells Fargo·Tech KnowledgeL3Hiring Manager Round

Explain LLM vs RAG, fine-tuning, and RAG pipeline optimisation

via2dbi

Tests broad ML/AI literacy and practical hands-on exposure.

LLM vs RAG: LLMs generate responses from parametric (baked-in) knowledge; RAG (Retrieval-Augmented Generation) augments generation with external retrieved documents, reducing hallucination and enabling up-to-date answers without retraining.
Fine-tuning: further trains a pre-trained model on domain-specific data to shift its parametric knowledge; costlier than RAG but useful when tone, format, or specialised reasoning is needed.
RAG pipeline optimisation: chunking strategy, embedding model choice, vector store tuning, re-ranking retrieved chunks, query expansion, hybrid search (dense + sparse), and context window management.
Layman's LLM: predict the most plausible next word/token at massive scale, producing fluent, context-aware text.

Candidates are also expected to share concrete AI tools they use (e.g., Copilot, Cursor, LangChain) and real use-cases, signalling practical familiarity rather than just theoretical knowledge.

Add a follow-up question they asked

When would you fine-tune instead of using RAG?

How do you optimise a RAG pipeline: chunking, embeddings, re-ranking?

How do you evaluate and reduce hallucinations in a RAG system?

asked …