IEEE Computational Intelligence Society Student Chapter, IIT
ModelX

Build the system that sees what people can't.

Expertise does not scale. So most people decide half‑blind.

Someone can always read the signals: what a price rise will do to a household, what a swelling river means for the village below it, why an illness is spreading where it is. There have never been enough of them.

For most people, most days, that expert is unavailable, unaffordable, or unreachable. They decide anyway, on whatever they can see.

ModelX 2.0  /  The Expertise Gap

01

Turn real data into insight a person could not have reached alone.

Three domains. Pick one, then find your own problem inside it. A real problem, affecting real people.

You choose the user. You find the data. Machine learning, deep learning, agentic systems, retrieval, anything. The method is yours.

Your system must tell its user something they did not already know: a pattern, a risk, a trend, an anomaly. Retrieval is not insight. The intelligence has to come from the data.

The hard part is producing an output worth trusting: knowing what the data cannot support, and failing safely when a decision depends on you.

02

Three domains. Pick one.

Each domain is wider than the examples below. Find your own problem inside one of them; we are not handing you a list to pick from.

Economy & livelihoods

Whole economies run on information most of the people inside them never get to see, and the cost of guessing wrong falls hardest on those who can least afford it.

How people and communities earn, spend, trade and get by. Work, wages, small business, informal economies, prices, rent, debt, markets, public money, and anything else in that space.

Tabular, time series, often refreshing daily

Climate, environment & disaster risk

The environment shifts around people faster than they can see it, and the warning usually arrives too late to act on.

The environment people live inside. Weather, water, air, land, heat, energy, hazards, and anything else in that space.

Geospatial, satellite, sensor networks

Health & wellbeing

Health is decided long before anyone reaches a clinic, by conditions most people never get to see.

Health at the level of populations, not patients. Disease, nutrition, access to care, mental health, sanitation, behaviour, and anything else in that space.

Non-clinical only. No diagnosis, no treatment advice. Systems about populations, access, patterns and behaviour are in scope. Systems that act as a doctor are not.

Epidemiological, survey, text

Two submissions. Two cuts. One final.

03

Dates and deadlines.

7 Sep to 6 Oct

Registration

7 to 9 Oct

Teams locked, no changes after this

Sat 10 Oct

ModelX 2.0 opens

10 to 19 Oct

Phase one: Research

Mon 19 Oct

Project proposal due

Submission 1
Thu 22 Oct

Results announced

Gate 1
Date to be announced

Workshop 1

20 Oct to 10 Nov

Phase two: Implementation

Date to be announced

Workshop 2

Tue 10 Nov

Implementation due

Submission 2
Mon 16 Nov

Fifteen finalists announced

Gate 2
Sat 21 Nov

The Final: live pitches, winners and awards

04

Two submissions.

Project proposal

Due Mon 19 Oct

The kind of proposal you would hand a client. Every section must be present and substantively answered.

The problem you have chosen inside your domain, who it affects, and the user you are building for. A review of what already exists and why it does not solve this for them. Your proposed solution and methodology. Your data: where it comes from, how you will extract it, whether it is historical, recent or live, and what right you have to use it. How you will judge whether it worked, and what you will do if it does not.

Implementation

Due Tue 10 Nov

What you actually built, and the code behind it.

A technical report covering the architecture, the stack you chose and why, and an honest evaluation against your proposal. An account of the data you finally used, its temporality and its gaps. A git repository anyone can open, with a README that gets it running. A two-page summary, and a three-minute video of the system working on real data.

Submit your implementation
05

How you are judged.

Implementation: cuts to 15

Data work30%

Sourcing, extraction, and above all temporality. Plus cleaning, honesty about gaps, ethics and licensing.

Insight delivered25%

Does it tell the user something they could not have reached alone?

Engineering quality25%

Structure, dependency handling, a git history showing sustained work, and a report that matches the code.

Functionality demonstrated20%

What your demo video actually shows working, on real data, end to end.

The Final: decides the winners

The system, working35%

Running live, on real data, in front of the panel. Demonstrated beats described. Includes how it handles what it doesn't know.

Insight and its value25%

Would your named user actually act on what it tells them?

Problem significance20%

How big a problem you chose to take on, and how many people it affects.

Depth under questioning20%

Whether you can defend your choices when pressed, and explain the work clearly and honestly.

06

Rules and eligibility.

Team size

One to four members, from any university, solo entry included. Each person may be on one team only.

Rosters lock 6 October

No additions, substitutions or swaps once registration closes. A team that loses a member continues with whoever remains.

Original work

Everything you submit must be your own, produced during the competition. Bringing a finished system, or plagiarism in code, documents or data, is disqualifying.

Open source is expected

Libraries, pretrained models and public datasets are freely usable. Credit them.

AI assistance is permitted

Use the tools. Disclose which ones and what for, and own what you submit. "The model wrote it" is not a defence.

Your data must be yours to use

Licence, terms of service, or documented permission. Respect robots.txt directives. Minimise and anonymise personal data. Data obtained through deception is disqualifying.

Deadlines are firm

Late submissions are not accepted, and a missed gate is an elimination. A private repository nobody can open counts as no submission.

Conduct

Harassment, discrimination or abuse means immediate removal. So does sabotage, interference with another team, misrepresenting your results, or any attempt to influence how your work is assessed.

Appeals

Within 48 hours of a result, on grounds of procedural error only, not disagreement with a score. The committee's decision is final.

07

What you're competing for.

Overall placings

01

Winner

Rs 75,000

02

First runner-up

Rs 50,000

03

Second runner-up

Rs 25,000

Special award

Best Data Vision

The clearest thinking about data: where you found it, how you got it, and above all whether it was current enough to be worth acting on. Judged from your implementation rather than your pitch, so every team that submitted one is in contention.

Every team that passes the proposal gate receives a certificate of participation. You do not have to reach the final fifteen, or win anything, to leave with recognition for the work.

08

Contact us.

Dinugi Nimnara

Dinugi Nimnara

Volunteer Management Lead

dinugi.20231363@iit.ac.lk 071 995 1662
Chanuka Wijeratna

Chanuka Wijeratna

Event Co‑chairperson

chanuka.20232021@iit.ac.lk 076 564 8877
Adrian Malcolm

Adrian Malcolm

Event Co‑chairperson

mellitus.20250440@iit.ac.lk 074 037 5521