Top Infosys AI Interview Questions — RAG & Agentic (2026) with clear, practical answers for 2026 interviews.
What is RAG and why does it matter?
Retrieval-Augmented Generation grounds an LLM with retrieved documents so it answers from your data instead of guessing — reducing hallucination and enabling citations. It is the default pattern for enterprise GenAI.
Walk through a RAG pipeline.
Ingest → chunk → embed → store in a vector DB → at query time embed the question, retrieve top-k, optionally re-rank, build a grounded prompt with citations, generate, and evaluate faithfulness.
How do you choose a chunking strategy?
Structure/semantic-aware chunks (by heading/paragraph) with small overlap, sized to the embedding model and question type. Too large dilutes retrieval; too small loses context.
What is re-ranking and why use it?
A second stage that reorders retrieved chunks by true relevance (e.g. a cross-encoder) before generation — usually a bigger accuracy win than swapping the LLM.
How do you reduce hallucination?
Ground strictly in retrieved context, cite sources, instruct the model to say "I don’t know", re-rank for relevance, and run groundedness evaluation.
How do you evaluate a RAG system?
Retrieval metrics (hit rate, MRR/nDCG) and answer metrics (faithfulness, relevance, completeness) on a curated eval set — iterate on chunking, embeddings and re-ranking.
Fine-tuning vs RAG?
RAG for fresh/changing knowledge and citations; fine-tuning for style, format or narrow behaviour. Often combine both.
What is Agentic AI?
LLMs that take actions through tools in a loop (reason → act → observe), not just answer questions. They plan, call tools, use memory and can run multi-step workflows.
Prepare with a deployed project from our AI, GenAI & Agentic AI, DevOps and cloud tracks, and see live roles on the jobs board. Book a free demo: +91 96660 19191.