Every roadmap for AI careers I see is a wall of forty technologies. Students screenshot it and freeze. The roadmap that got me from a BCA classroom in Kanpur to building production AI systems had five stages and one rule: at every stage, ship something you can show. Here is the twelve-month version I give students who ask.

Months 1–2: Python, APIs and one deployed app

Python well enough to read library code. HTTP, JSON and calling an LLM API. Then a small web app with FastAPI and a front end, deployed with a URL. Not an AI app yet — a working one. This is the floor everything else stands on.

Months 3–4: a RAG project with evaluation

Chunking, embeddings, hybrid retrieval, citations and a confidence threshold, over a corpus you care about — the college-documents project is ideal. Write a fifty-question evaluation set and publish the numbers. This single project answers most GenAI interview questions.

Months 5–6: a fine-tuning project

QLoRA on a small model on a free GPU, with a validated dataset and a base-versus-tuned comparison. Learn what changed and what did not. Now you can speak to both halves of the field with evidence.

Months 7–9: ship for a real user

Take one project to ten real users — a club, an NGO, a shop. Add logging, error handling, a fallback, a privacy note. Enter one hackathon with the same discipline. Real users are what separate a portfolio from a repository.

Months 10–12: make it visible and apply

What to ignore

Certificates as a goal, courses without a project, and the urge to learn every framework before shipping one thing. The students who got roles were not the ones who knew the most tools; they were the ones who could show a system working and explain why it was built that way.

— Pranjul Rathour, GenAI Engineer from Kanpur, India. Open to GenAI roles, hackathon judging, mentorship sessions and guest talks at any campus: pranjulrathour41@gmail.com.