The problem HackMate AI — AI-Powered Hackathon Team Builder solves

Hackathon teams are usually formed by chance — whoever you happen to sit next to.
That leads to overlapping skills, missing roles (no one covering design, or no one
covering ML), and hours lost negotiating groups instead of building.

HackMate AI fixes this. Participants register their skills, interests, role,
experience, availability, working style, and a short bio. The system uses NLP
embeddings and skill-complementarity scoring to evaluate how well people's skills
fit together, then an optimizer generates balanced teams — each with a skill
coverage score, a compatibility score, and a plain-language explanation of why
that group works. Participants can also use "Find My Teammates" to get ranked,
individually-explained matches instead of a full team.

Challenges we ran into

Turning "skills on paper" into a fair team split was harder than expected. Two
people can list the same skill tag but mean very different levels of depth, so
naive keyword matching gave misleading compatibility scores. We had to move
toward embedding-based similarity and tune the balance between skill coverage
and shared interest so the optimizer didn't just cluster people with identical
skills together. Getting the explanation text to stay accurate to the actual
scores — rather than generic filler — took a few iterations too.