AI Attendee Matchmaking

How EventIntro's AI matchmaking works: a five-question survey, LLM profile enrichment, seek/offer keyword embeddings, and complementary nearest-neighbour search.

Who this is for

Anyone evaluating how EventIntro actually decides who meets whom.
Buyers comparing real matchmaking against keyword-tag "matching".
Technical evaluators who want the pipeline described plainly.

How does EventIntro's matchmaking pipeline work?

Five steps. An attendee answers a five-question survey; an LLM expands those answers into a fuller profile and extracts what the person is seeking and offering; those seek/offer keywords become vector embeddings; a nearest-neighbour search finds the people whose offers answer this attendee's needs; and a group-formation step assembles balanced breakouts. The output is a ranked, reasoned match list per person.

Every stage exists to convert vague human intent into something a machine can compare. The survey captures intent, the LLM enriches it, the embeddings make it comparable, and the search turns comparison into specific introductions.

How is this different from keyword-tag matching?

Tag matching asks attendees to self-select from a checklist and pairs on overlap — shallow, gameable, and blind to anything not on the list. EventIntro matches on meaning: embeddings place semantically related needs and offers near each other even when the words differ, so "needs distribution" finds "runs a newsletter" without anyone having tagged either.

Which model runs it, and can humans override?

It supports Gemini, OpenAI, Anthropic, and Groq with automatic fallback; match quality depends more on the survey and embedding design than on the model. And every match is previewed and adjustable before an event — the AI is a strong default, not a locked verdict.

Frequently asked questions

Is this real matching or just keyword tags?

Real matching. Tag systems ask attendees to pick from a list and pair on overlap, which is shallow and gameable. EventIntro runs survey answers through an LLM to build a richer profile, extracts what each person seeks and offers, embeds those as vectors, and finds complementary fits by nearest-neighbour search — matching meaning, not labels.

Which LLM does it use?

Multiple providers — Google Gemini, OpenAI, Anthropic, and Groq — with automatic fallback. Match quality depends far more on the survey design and embedding pipeline than on which model handles a given request.

Can a human override the AI?

Yes. Every match is previewed and adjustable before an event goes live. The AI is a strong default, not a final authority — the facilitator keeps the last word.

Thanks — we've got it.
We read every one of these. We'll be in touch at the address you gave us.