GT Swarm

Find matches

Find your people, or find your startup.

Go beyond keywords. GT Swarm connects startup needs with relevant projects, skills and engineering experience across Georgia Tech. Matches combine technical experience, semantic relevance, collaboration preferences and documented Georgia Tech connections.

Role I want
Or fill an example:

How the intelligence works

Matches come from what people have built, and every one explains itself.

  • Project and experience match

    55%

    Your request is compared with each person’s projects, bio and solved problems: 30% keyword relevance (exact terms like BLE or nRF52) and 70% Gemini semantic relevance (the same work described in different words). We picked the split by testing 15 queries with hand-written relevance judgments.

  • Documented experience

    20%

    How much of their record sits among the closest projects and solved problems.

  • Skill relevance

    shown

    Skills you asked for that are on their record are highlighted on every card.

  • Role eligibility

    filter + 10%

    Pick a role and only people open to it are ranked. No one is shown for a role they did not choose.

  • Shared connections

    10%

    Same projects, teams or majors, and accepted connections, from the knowledge graph.

  • Availability

    5%

    Hours per week people say they can give.

Grounded explanations. Each reason is built only from stored records (a project, a solved problem, a stated role), never invented. Click Why this match? on any result to see how its score adds up. Lists only show clear matches; weaker ones stay one click away.

The other direction: startups are recommended to a builder by text similarity (35%), skills the startup needs that you have (25%), a role it needs that you are open to (20%), interests (10%), stage (5%) and shared ties (5%).

builtusesknowssolvedonAlumStudentProjectSkillProblem
A knowledge graph links people, projects, skills and solved problems. It powers the shared-connection score, warm introduction paths and the network maps.

Gemini embeddings (gemini-embedding-001) are precomputed once per record and cached; a search embeds only the query. If Gemini is slow or unavailable, matching falls back to keyword search within a few seconds, and nothing else changes. Semantic similarity never draws a connection line and is never treated as proof of a skill. Get Unstuck uses the same graph to find who can help with a problem.