Module 8 · Safe, fair, and your career · Lesson 31 of 32
GenAI careers and interview questions
11 minFree GenAI course · BeginnerLesson 8.2By Saket Bhatnagar
Researched and drafted with AI, reviewed and edited by Saket Bhatnagar.
In short
- GenAI work is a team sport: AI engineers, ML engineers, data people, product managers, designers and evaluation specialists.
- Most people start by building small, real projects on top of existing models, not by training models from scratch.
- Interviews test whether you can explain tokens, RAG, agents, evals and safety in plain words, with a trade-off and a way to test it.
Video lesson · coming soon
GenAI careers and interview questions, honestly explained
Lala, your AI mentor, will walk you through this lesson in a short animated video. Until then, everything is here to read and try.
- 01 Five AI jobs. Which one is you?
- 02 GenAI work is like a big wedding
- 03 What they really do all day
- 04 A realistic way to start
- 05 Questions interviewers really ask
- 06 No shortcuts, but a clear path
- 07 Practise, then build your capstone
Arjun is in his final year in Nagpur. The placement cell posts five openings, and every one has “AI” in the title: AI Engineer, ML Engineer, GenAI Developer, Data Analyst (AI), AI Product Associate. He doesn’t know which one is him, or whether any of them is.
If you feel like Arjun, this lesson is your map. You will see what each role does on an ordinary Tuesday and a realistic way to begin. Then come a dozen questions interviewers really ask, with short answers you now understand.
Which jobs actually use GenAI?
Titles are messy. One company’s “GenAI Developer” is another’s “AI Engineer”, and some “ML Engineer” jobs are really app building. Read the job description, not only the title.
| Role | What they build or decide | Core skills | A common way in |
|---|---|---|---|
| AI engineer | Chatbots, RAG search, agents, AI features in apps | Programming, APIs, prompting, RAG, evals, cost | Software developers adding GenAI projects |
| ML engineer | Training, fine-tuning, serving and monitoring models | Python, maths and statistics, deep learning frameworks, GPUs | Strong CS or maths background, ML projects, often a master’s |
| Data engineer or analyst | Clean data pipelines, document stores, dashboards on AI use | SQL, Python, data pipelines, analysis | Data roles, now using AI to clean, label and query data |
| AI product manager | Which problems deserve AI, what success means, launch plans | User research, writing, metrics, AI basics, judgement | PMs, analysts or engineers who understand users |
| Designer for AI | How people ask, wait, trust, correct and undo AI | UX design, writing, prototyping, research | UX designers learning AI’s strengths and limits |
| Prompt and eval specialist | Prompts, test sets, rating guides, red-team attacks | Clear writing, domain knowledge, careful testing | Writers, testers, teachers and domain experts |
What should you learn first?
Good news: you have already learned the ideas behind most GenAI interviews in this course. What remains depends on the role you pick.
- For AI engineering: one programming language well (Python or JavaScript are common), how web APIs and JSON work, and basic databases. Add the course topics of prompting, RAG, tools, cost and evals.
- For ML engineering: all of the above, plus linear algebra, probability, statistics and a deep learning framework. Expect to study longer.
- For data roles: SQL and Python first, then how documents get cleaned, chunked and indexed for RAG.
- For product, design and eval roles: clear writing, talking to users, and knowing what AI does well and badly. Hands-on use of AI tools matters more than code.
How do you get started, realistically?
Think of learning to cook. You don’t open a restaurant first. You make dal for your family, notice it was salty, fix it, and make it again. GenAI careers grow the same way, through small real dishes.
A realistic path into GenAI work
Drag to exploreEach step produces proof you can show. The loop from feedback back to building is where most of the learning happens.
Diagram connections:
- Learn the basics to Build one small helper (start small)
- Build one small helper to Test it honestly (measure)
- Test it honestly to Write it up (explain)
- Write it up to Practise interviews
- Practise interviews to Apply and get feedback
- Apply and get feedback to Build one small helper (feedback)
- Learn the basics: This course, plus one programming language or your own fiel…
- Build one small helper: Solve a real problem for real people: a class, a shop, a so…
- Test it honestly: A test set, a cost estimate and a list of what failed.
- Write it up: A short page or README people can read in five minutes.
- Practise interviews: Say your answers aloud until they sound like you.
- Apply and get feedback: Rejections tell you what to build or study next.
Pick a problem you can touch
Choose something with real users near you: a college fest help desk, a hostel mess menu bot, a kirana shop’s stock questions, notes Q&A for your class.
Build the smallest version
One prompt, a few documents, maybe one tool. A working small thing beats an unfinished big one.
Add tests and a cost note
Write twenty test questions with good answers, score your bot, and estimate the monthly bill. Very few beginner projects do this, so yours will stand out.
Write down what failed
Interviewers love “it hallucinated on dates, so I added the day to every chunk”. Failures you fixed are your best stories.
Use your current job
If you already work as a tester, analyst, support engineer or teacher, bring GenAI into that work. Moving sideways inside a company is often simpler than a cold start.
Be patient with yourself. Entry-level AI roles get many applicants, and a lot of good people take months to land one. Steady building, testing and practising is what moves the needle, not a single course.
What do GenAI interviews ask?
Most GenAI interviews mix three things: concept questions, a “design this AI feature” question, and questions about your own project. Here are twelve concept questions that come up again and again, with short model answers.
| Question | Short model answer | Lesson to revisit |
|---|---|---|
| What is a token, and why does it matter? | A small piece of text the model reads as a number. Cost, speed and context limits are all counted in tokens, and Indian languages often use more tokens per word. | Tokens: how AI reads your words |
| How does an LLM generate an answer? | It predicts a probability for every possible next token, picks one, adds it, and repeats until it stops. | Next-word prediction: the one trick behind chat AI |
| What does temperature control? | How adventurous the token choice is. Low gives safer, more repeatable answers; high gives more variety and more risk. | Temperature: the creativity dial |
| What is a context window? | The most tokens the model can see at once, counting instructions, history, documents and the reply. Anything outside it is invisible to the model. | Context window: how much AI can keep in mind |
| Why do LLMs hallucinate, and how do you reduce it? | They are trained to produce likely text, not checked facts. Reduce it with RAG, “say I don’t know” instructions, citations and tests. | Why AI makes things up (hallucinations) |
| Explain RAG. | Search your documents for chunks related to the question, put them in the prompt, and ask the model to answer only from them. | RAG: the open-book exam |
| What are embeddings and vector search? | Embeddings are lists of numbers that place text on a map of meaning. Vector search finds the chunks closest in meaning to the question. | Embeddings: meaning as a map |
| RAG or fine-tuning? | RAG for knowledge that changes or must be cited. Fine-tuning for style, format or narrow habits. Try prompting first. | Prompting, RAG or fine-tuning? |
| What is tool calling? | The model writes a structured request, such as “get_weather for Pune”; the app runs it and gives the result back for the final answer. | Tools: giving AI a phone and a calculator |
| What is an AI agent, and what are its risks? | A model in a loop of plan, act, observe. Risks: wrong actions, runaway cost and prompt injection; add step limits and human approval. | Agents: think, act, check, repeat |
| How would you evaluate an LLM feature? | Build a test set of real questions with good answers, score every change against it, include tricky and unsafe cases, and watch live use. | How to know your AI is good |
| How would you cut cost and latency? | Shorter prompts, caching repeated parts, smaller models for simple jobs, limits on answer length, and streaming the reply. | Cost and speed: the token meter |
For design questions, like “design a support bot for a food delivery app”, use the capstone in the next lesson. Walk through the same parts: users, prompt, documents, tools, cost, tests and safety. Interviewers care more about your reasoning than about naming any one product.
Check yourself
Check yourself
Priya is a backend developer with two years of experience and wants to move into GenAI work. What is the most direct first step?
Pick an answer to see why it is right or wrong.
Check yourself
An interviewer asks: “When would you fine-tune instead of using RAG?” Which answer is strongest?
Pick an answer to see why it is right or wrong.
Check yourself
Who usually decides whether a feature should use AI at all, and how success will be measured?
Pick an answer to see why it is right or wrong.
Remember these three
- GenAI is a team of roles; most beginners enter through AI engineering, data, product, design or evals, not model training.
- A small, tested project with an honest write-up is the strongest proof you can show.
- Answer interview questions in four parts: define, example, trade-off, how you would test it.
Questions people ask
- Do I need a degree in AI to get a generative AI job?
- Not for most roles. AI engineering, product, design and eval work value skills and projects over a specific degree. Research and some ML engineering roles often prefer a strong maths background or a master’s.
- Is prompt engineering a real career?
- The skill is real and widely used, but fewer job ads use “prompt engineer” as a standalone title than a few years ago. Today it usually sits inside AI engineering, evaluation, content and product roles.
- Which programming language is best for generative AI?
- Python is the most common for ML and data work. JavaScript or TypeScript is common for AI features in web apps. Pick one, learn it well, and the ideas carry over.
- How long does it take to become an AI engineer?
- It depends on where you start. A working developer can build credible GenAI projects within a few months of steady practice; a beginner to programming needs longer. Nobody can promise a timeline to a job.