Skip to content
Tech Interview Prep

Real Interview Questions for the Agentic AI Engineer at Tata Consultancy Services (TCS)

New

Overview

How the loop runs

Candidates describe several different shapes, so treat the format as unsettled. One walk-in drive in Hyderabad ran five rounds: an aptitude and coding assessment, a technical HR round, a general HR round, then document verification and onboarding for those who cleared. Another candidate, interviewing for a Gen AI role, described a conversation led by the resume: a detailed walkthrough of past projects, questions on RAG, Azure AI Search, re-rankers, agents and OpenAI models, then writing a RAG implementation in Python. A digital interview in Chennai started with a self-introduction, moved through surface-level AI concepts (LangChain versus LangGraph, agents, RAG), and included a Fibonacci question that the candidate solved in Java after the interviewer allowed it. A GeeksforGeeks account of an on-campus AI/ML fresher interview began with the LangChain, LangGraph and LangSmith projects on the resume, then covered Agentic AI, CrewAI and FastAPI, a retail revenue scenario, and a printed code snippet to explain.

The reports are few and anecdotal. One Interview Query source says all three of its candidates received no offer. Its reports also disagree on depth: in one, two easy coding questions were followed by a software engineering question that ended the interview after a stumble. Expect range rather than a fixed script.

What interviewers are evaluating

Across the accounts, interviewers check whether you can explain systems you have built, and whether you can defend the design when pushed. Candidates describe being moved from architecture to working code, and from a project summary to the reasons behind each choice. Basic Python fluency and core CS fundamentals come up early in the walk-in and digital reports, and a weak answer there can end the conversation before the AI questions start. The judgement being bought is whether you can reason about where a retrieval pipeline fails, and whether the claims on your resume hold up under follow-up.

Three focus areas

  1. Retrieval pipeline design. Ingestion, chunking large documents, measuring retrieval quality with Recall@K and Precision@K, and enforcing document access controls before retrieval. The LinkedIn account and the Gen AI report both return to this area.

  2. Controlling LLM output. Temperature, Top-K and Top-P, forcing JSON or CSV, and function calling. Expect questions on how the model picks a tool and how you handle a malformed or partial tool response.

  3. Agent frameworks and project depth. What Agentic AI means, when a framework such as CrewAI or LangGraph fits, and a walkthrough of your own projects. Two reports say the project discussion took most of the interview time.

Basic coding runs alongside these in every report, so Python fundamentals and a clean Fibonacci or small array problem are worth keeping sharp.

How to prepare

Practise answers aloud and time them at two to four minutes. Lead with the direct answer, then cover the trade-offs and failure modes an interviewer would follow up on. Work through the Interview Q&A on this card for the RAG, sampling and function-calling questions, then take the Practice MCQ to check that you can choose between options once a constraint is stated. Before the interview, write a one-page walkthrough of each resume project, including the stack named on it, since reports say the interviewer starts there. Use the role's own prep for the Top 100 and concept roadmap. Finally, write a small RAG pipeline in plain Python from memory, because a candidate in one report was asked for exactly that.

Feedback on this overview

Curated: · Written: · Reviewed: