A redesign of what an AI programme for engineering colleges should look like in 2026: shorter, build-first, artifact-driven, and deliverable by one instructor. An initiative by IIT alumni, for Indian engineering colleges. To take the guesswork out of building AI applications to every nook and corner of our country.
Old design: teach AI and data concepts, then build something at the end.
New design: start building in hour two, and teach a concept only at the moment the build breaks without it.
Every track runs one project for the whole programme. Not a new mini-project weekly — one thing, built in visible weekly increments, that the student owns and can defend.
Captains, self-validating check.py, pre-recorded concept blocks, fixed environments, and rolling vivas make 100+ students survivable solo.
Browser-first for weeks 1–2. Pre-cached models on USB and Drive. One full track runs offline after day 1. MOCK mode when APIs die.
2–6 week units fit clean windows without crossing blackout periods. You can run 6–8 cohorts a year instead of two.
Python, Colab, Streamlit Community Cloud, free Gemini tier, open-source embeddings. Verified free/open stack, August 2026.
Baseline / Plus / Edge ladder inside one project. Captains from the top 20%. Nobody bored, nobody left behind.
100-point rubric. 40 marks automated. Viva about their code only. Integrity rule: if they can't explain it, we cut scope and get an honest project.
The original package had unusually strong pitch playbooks, college-specific openings, academic-calendar mapping, and an objection bank. Those survive. What changed is the product being sold.
A good sales machine attached to the wrong product has become a 4-week build programme that one person can run and that produces something a recruiter can click on. The viva was already the best idea. Everything here is built to make the viva worth passing.
Start a conversation