Hackathon Saathi

Why it needed to exist

At hackathons, teams usually form by chance or friendship. Teams end up with skill gaps or clashing working styles, and many fall apart before the deadline.

Hackathon participants looking for teammates, and organisers who want more teams to finish strong.

What it does

• Skill-balanced teamsRecommends teammates who fill the gaps in your skills rather than duplicating them.
• Collaboration fitUses psychometric indicators so teams are emotionally stable, diverse and ready to work together.
• Works for newcomersA cold-start tier gives good recommendations even to someone with no history.
• Gets better over timePersonalised recommendations improve as the system learns from outcomes.

What I did

I worked on the recommendation logic and the research that validated it.

How it works

1. Recommend

   Hybrid engine

   Content-based and collaborative filtering, matrix factorisation and reinforcement learning.

2. →

   Tiers

   Cold-start → global → personal

   Picks the right strategy depending on how much the system knows about you.

3. →

   Optimise

   CP-SAT model

   Psychometric indicators become constraints, so the final team is balanced as a whole.

Choices that shaped it

Optimise the whole team, not pairs

Great pairs can still make a poor team. A constraint solver balances everyone at once.

Treat working style as a constraint

Skills alone don't predict whether a team finishes. Psychometric fit is built into the model, not bolted on.

Three tiers for cold start

New users get sensible matches immediately instead of waiting for data.

Where it stands

The matching engine is complete and achieved a 100% success rate in our in-house research. The end-to-end web app isn't finished, so it isn't publicly hosted yet.

Next: Finish the end-to-end web app; Pilot it at a live hackathon and feed real outcomes back into the model.