Baseline
No eliminationA short round on ML fundamentals and experiment selection, to see where everyone is starting from. Nobody is cut.
Reverse Engineer the Intelligence
You are not given a dataset. You are given a system that answers questions — and a limited number of questions to ask it. Everything else about it is hidden: the data, the features, the pipeline, the model, the thresholds. Your job is to work out what is inside.
A toy black box. Move an input, spend a query, read the output.
You are not told what it computes. Working that out is the entire competition.
A conventional hackathon runs dataset → train model → submit accuracy. BLACKBOX AI runs the other way round. Your team is given access to a private machine learning system. You can send it inputs and read its outputs. Nothing else.
You do not know the training data, the preprocessing, the engineered features, the algorithm, or the decision thresholds. You have a fixed budget of queries, so every question you ask has to be worth asking.
Six steps, run over and over, under a shrinking budget. The teams that win are the ones whose fifth query is chosen because of what the first four returned.
Back to 01, with fewer queries left
A test of experimental reasoning under uncertainty — designing the experiment that separates two competing explanations, not memorising which library has which default.
After a short baseline round, each stage hands you a hidden system and a query budget, and each one goes deeper than the last. Full rules and timings are given at the event briefing.
A short round on ML fundamentals and experiment selection, to see where everyone is starting from. Nobody is cut.
Your first real black box. Work out which features matter, where the thresholds sit, what is stable and what is not. Prove what you claim.
Past surface behaviour into hidden preprocessing and feature interactions. By now several explanations fit what you have seen — find the experiment that tells them apart.
Find where the system fails. Edge cases, unstable boundaries, systematic failure modes — with reproducible evidence for every weakness you claim.
Build your own model that reproduces the hidden system's behaviour. Your queries are your training set. Judged on what your model captures, and how honestly you tested it.
A fresh hidden system, identical for every finalist. Everything at once — then you explain your reasoning to the judging panel. Understanding has to back up the leaderboard.
IEEE Day marks the first time engineers worldwide gathered to share their ideas, in 1884. It falls on the first Tuesday of October — which in 2026 is 6 October, the opening morning of BLACKBOX AI 1.0.
It is marked every year by student branches and sections around the world — talks, competitions, outreach, whatever each branch chooses to run. The point is the same everywhere: engineers doing something together rather than separately.
The IEEE Student Branch at GCET is marking it alongside the competition, so Day 1 carries both — the IEEE Day celebration and the opening rounds of the event.
“Leveraging Technology for a Better Tomorrow”
Certificates for all participants. Best ML Investigator recognises the clearest experimental reasoning of the event — the team whose investigation best shows how they thought, independent of where they finished.
College policy requires students to obtain prior permission from their HOD before participating in or organising events. Students who participate without prior HOD permission will not be granted attendance. Please arrange this before 6 October.
Anything not answered here goes to the organisers through the WhatsApp group you join after registering.
No, but you should be comfortable with the basics — training a model, reading a scatter plot, reasoning about features. Round 0 is deliberately a baseline round that nobody is eliminated on. The later rounds reward careful thinking more than they reward knowing an exotic algorithm.
Yes, genuinely. We are not trying to run an AI-free competition — that is neither realistic nor useful. The difficulty comes from somewhere an assistant cannot reach: the hidden systems are private, built by us, and never exposed. An AI can help you analyse what the black box told you. It cannot tell you what the black box will say.
Two or three students. One account per team, which all members can be logged into at once. The query budget is shared across the team, so coordinate — two people running the same sweep spends it twice.
A laptop with Python and your usual ML libraries already installed and working — set this up before you arrive, not on the morning. Bring your charger. Details on connecting to the competition network are given in the Day 1 briefing and in the WhatsApp group.
Round 1 results are published on the evening of Day 1 precisely so that you know before Day 2 begins. If your team does not qualify you are still welcome to attend, and you keep your participation certificate.
Yes, entirely free. Sign up through the form to reserve your place and join the WhatsApp group — that group carries all official announcements. You create your actual team account at the venue on Day 1, and that is where your team name and login come from.
IEEE — the Institute of Electrical and Electronics Engineers — is the world's largest technical professional organisation, with hundreds of thousands of members across more than 160 countries. Its roots go back to 1884; the modern body formed in 1963.
Most engineers meet IEEE long before they join it, usually without noticing:
Every float your model trains on follows IEEE 754. Every packet in this room follows 802.11 or 802.3.
IEEE's societies span power and energy, communications, robotics and automation, signal processing, biomedical engineering, aerospace, control systems, photonics and more — around forty of them. Whatever you end up building, there is almost certainly a society, a conference and a body of literature for it.
Our student branch sits with the IEEE Computer Society, which is where this event comes from — but membership is not limited to that.
IEEE Xplore holds several million technical documents — papers, standards, conference proceedings. Being able to read the current work in your area, rather than a summary of it, changes how you approach a problem.
Student rates for conferences, and a real route to presenting or publishing your own work while still an undergraduate.
Join the societies that match what you care about. Each brings its own publications, competitions and community — and they are not restricted to your degree title.
Student branches are run by students. Organising an event like this one is itself the benefit — it is experience you cannot get from coursework, and it is visible on a CV in a way attendance is not.
Local sections, regional events and a global membership. The useful part of a professional body is usually the person it puts you next to.
Funding, awards and student competitions run at branch, section and regional level throughout the year.
This is the part most students miss while they are still in college, and the part that ends up mattering most.
Your college's journal access ends the day you leave. IEEE access does not. For anyone who keeps reading — and in this field you have to — that gap is the difference between staying current and guessing.
A student branch is your peers. The local section is not: it is people ten and twenty years into the career you are starting. That is a different room to be in, and membership is how you get into it.
IEEE is recognised internationally. For graduate applications abroad, or a first job in another country, publications and a professional grade are legible to people who have never heard of your college.
Member, then Senior Member, then Fellow — each earned, each meaning something specific to people in the field. It is a track you can actually be on rather than a badge you buy once.
Not the careers page. The person who reads your poster and asks a good question about it is often the person who remembers you a year later.
The 802.11 in your laptop was written by people in working groups anyone can participate in. Contributing to a standard is engineering with a longer half life than most products.
A subscription is something you pay for and then receive. This is the other way round: almost everything worth having from it — the society you join, the people you meet, the paper you present, the event you run — happens because you did something, not because you paid. The fee buys you the door, not the room.
The Computer Society is one of its largest, and its societies cover robotics, biomedical, aerospace, power, signal processing and more. Your branch is what you make of it, not what your degree is called.
Publishing is one route. Reading the current work in your area, joining a society, running events, and meeting people outside your college are the parts most students actually use.
It is, if you only pay the fee. What is worth something is what you did with it — organised, presented, competed, built. That is visible; membership alone is not.
No. First years can join, and branches usually need the people who show up more than they need the people with the highest marks.
A branch is only as active as the students running it — there is nobody else to do it. This event was built by the branch, in our own college, for our own students. That is what happens when people show up.
Student membership is discounted, and joining runs through ieee.org. If you want to know what it is like in practice, ask any of the organisers on the day — they are all student members, and this event is what a branch does.
Register through the Google Form, then join the WhatsApp group linked at the end of it. All official announcements go through that group.
Registration has closed and the competition is under way at GCET. Announcements, round timings and results go out through the WhatsApp group. Standings are published on the Standings page as each round is evaluated.
BLACKBOX AI 1.0 ran on 6–7 October 2026. Thank you to everyone who competed, and to the judges who sat through the technical defences. Final standings are on the Standings page. A future edition, BLACKBOX AI 2.0, will introduce deep learning.