Chat prep

GenAI role prep notes

GlanceOne line each.
Skipwhat you already know.
Tapa point. Type notes in its box.
Asked 0 of 12

Points

Yellow line = the point. Below = what others saw.

1All the steps, in orderTest → bot interview → tech → HR?

Ask

  • What are all the steps, start to offer?
  • Is there still an AI bot interview and a voice test?

What others saw real reports 2025

Online testAptitude, English, coding, tech MCQs, voice test
AI bot interviewA bot asks about your resume, then 1 easy code
Tech roundA person. Projects, Python, SQL, GenAI
HRAlways asks: OK to relocate? OK with shifts?
2What's in the online testSections, time, which website

Ask

  • Which sections? How long? Any cut-off?
  • Coding in it? Python or SQL?

What others saw real reports

  • A 2025 data-scientist test had many GenAI MCQs.
  • Voice test = speak English into the mic.
MCQ = pick one answer from fourAptitude = maths + logic puzzles
3All the topics they askNot only Python, SQL, FastAPI, GenAI

Ask

  • For each topic below: must, maybe or no?
  • FastAPI and MCP: for everyone, or only your team?
  • Will they ask from my degree (ML, stats)?

Possible topics

Pythonbasics, OOP, memory
SQLjoins, group by
FastAPIbuild an API
GenAILLM, prompts
RAGsearch docs, then answer
MCP · AgentsAI using tools
ML + statsfrom your degree
Azure OpenAIMicrosoft's AI cloud
Git · Dockersave code, pack app

Note our read

One LTM role was named "GenAI-Azure", so Azure tools may matter. No public report mentions FastAPI or MCP for freshers, so the list may be newer.

4How deep the answers goMeaning, inside working, or full design?

Ask

  • Up to which level below do freshers need to go?
  • One answer that got someone selected, one that got rejected.

Three levels, using RAG as the example

L1Meaning: "The AI first searches our documents, then answers."
L2Inside: split docs into chunks, turn them into numbers, pick the top 3 matches.
L3Design: draw the boxes, and say why this choice and not another.
5Design questions"Design a chatbot / system" – get 2 examples

Ask

  • Do freshers get design questions? How big?
  • 2 real examples, and what a good answer had.

What a design answer looks like

User API Searchdocs AI model Answer Every step is saved in a log, to find problems later

Questions that may come our guess

  • Chatbot that answers bank customers from policy PDFs
  • HR assistant: policy answers + apply for leave
  • Search 10,000 documents: store, search, update
  • Read bills and pull out amount, date, vendor
  • "Ask the database in English": how to stop it deleting data?
  • Give a chatbot memory of past chats
  • Serve an AI through FastAPI to 1000 users
  • Which AI model would you pick, and why?
HLD = the big boxes of a system and how they talkAPI = a door other apps use to talk to yours
6"What if" questions"AI gives a wrong answer…" – get 2 examples

Ask

  • 2 real examples, and what a good answer had.
  • Do they expect real-job thinking from a fresher?

Likely ones + direction of a good answer our guess

AI makes up an answerAnswer only from documents, show the source, test it.
Reply takes 10 secondsShow words as they come, reuse repeat answers, smaller model.
1000 users at onceMore servers, a waiting line, a limit per user.
The AI service goes downTry again, switch to a backup model, show a polite message.
The bill is too highReuse answers, shorter prompts, cheaper model for easy questions.
User types "ignore your rules"Safety checks on input; the AI never gets direct access to data.
Shows one customer's data to anotherCheck who is logged in before searching.
PDF is a scanned photoUse OCR first (reads text from images).
Is the new model better?Run the same test questions on both; compare scores.
Some users get an errorRead the logs to find the pattern.
Production = the live app real users useLatency = waiting timeHallucination = AI makes things up
7Coding roundLive or explain? Easy or DSA?

Ask

  • Live coding, or only explain? Paper, laptop or website?
  • Python, SQL or FastAPI? Can I use AI tools?

What others got real reports

EasyPrime number, factorial, reverse a string, remove duplicates
SQL2nd highest salary, joins, DELETE vs TRUNCATE vs DROP
NextTwo-sum with a dictionary, a decorator, a small FastAPI endpoint
DSA = puzzle-type coding problems
8Projects that impressSpend the most time here

Ask

  • What problems do their bank clients face every day?
  • What makes a project impress them? What do they ask about it?
  • GitHub links. RAG PDF again.

What stands out interviewers online

  • A real user, even a few
  • Messy, real-looking data
  • Test scores for the AI's answers
  • Numbers: cost and speed per question
  • One thing that failed, and how it was fixed
  • A person checks the AI before anything important
Deployed = live online, a link anyone can open Evals = fixed test questions that score the AI Metrics = numbers that prove it works Guardrails = rules that stop unsafe answers

They will ask

  • Why this problem? Who uses it?
  • Draw how it works.
  • How did you test it? What score?
  • What broke, and how did you fix it?
  • Which part did you build?
9ResumeWhat first? 1 page?

Ask

  • Projects first or skills first? 1 page?
  • Any good format? Can someone review mine?

Common advice blogs, not recruiters

  • 1 page
  • Projects with numbers, plus live link and GitHub
  • Skill words they search for: Python, FastAPI, RAG, LangChain, Azure OpenAI
10Why people get rejectedTop 3

Ask

  • Top 3 reasons people fail?

Seen online real reports

  • Weak Python basics
  • Can't explain their own project
  • Stuck on RAG basics (chunks, top-k)
  • Can't check code that AI wrote
11CertificateAWS or Azure – does it matter?

Ask

  • AWS Cloud Practitioner, or Azure AI-901 / AI-103?
  • Does it change who gets shortlisted?

Found so far real reports

  • No proof a cert changes shortlisting.
  • Two live projects count for more.
  • One LTM job post: certs "preferred".
12The real daily workGenAI building or support? Pune? Office?

Ask

  • Do new joiners really build GenAI, or do support / testing first?
  • Pune? Office every day? When does joining usually happen?

Why

  • Some "developer" jobs turn out to be data entry. Check before spending weeks on prep.

Study map

For later, not for the café. Tap to open.

Theory basics

Python · OOP · SQL · API · ML
Python
list vs tuple · list, set, dict · mutable vs immutable · shallow vs deep copy · *args, **kwargs · decorator · generator · == vs is · lambda · errors (try/except) · GIL · threads vs processes vs async · memory
OOP
class vs object · encapsulation, inheritance, polymorphism, abstraction · __init__ and self · overriding · abstract class
SQL
primary and foreign key · JOIN types · WHERE vs HAVING · GROUP BY · index · DELETE vs TRUNCATE vs DROP · window functions
API + FastAPI
what an API is · GET, POST, PUT, DELETE · status codes 200, 404, 500 · JSON · FastAPI vs Flask · Pydantic (checks input) · async def · auto docs page
ML (her degree)
supervised vs unsupervised · overfitting · train/test split · precision vs recall · regression vs classification · K-Means
Tools
Git basics · Docker (packs the app to run anywhere) · deploying an app online

GenAI map

Everything, not only RAG
LLM basics
token (a piece of a word) · context window (how much text it can read at once) · temperature (how creative) · embeddings (text turned into numbers by meaning) · hallucination
Prompting
zero-shot, few-shot (give examples) · step-by-step thinking · system prompt (the AI's standing rules) · answers in JSON
RAG
chunking (cut docs into pieces) · vector DB (stores the numbers: FAISS, Chroma, pgvector) · top-k (best k matches) · reranking · RAG vs fine-tuning
Fine-tuning
teaching a model with your own examples · LoRA (cheap way to do it) · when it is worth it
Agents + MCP
tool calling (AI asks to run a function) · agent (AI that plans steps) · MCP (a standard plug for tools) · LangGraph
Evals + guardrails
test set · faithfulness (answer matches the source) · AI-as-judge · RAGAS · prompt injection · hiding personal data
Running it live
speed · cost · caching · streaming · rate limits · logging · Azure OpenAI, AWS Bedrock

Project ideas to float

Would this impress?
Bank
Customer help bot that answers from bank papers and checks account status · loan-paper reader that flags missing documents · complaint sorter that marks urgency and drafts a reply
HR
Policy assistant for questions asked again and again · resume shortlister that gives reasons · new-joiner helper

Starting ideas only. Final picks come after more research.

Made 13 Nov 2026. "Real reports" = candidates' posts from 2025–26. "Our guess" = not confirmed yet.