How we built AI networking for the 4,500-person Moscow Startup Summit

Each group at the summit has its own goal. A startup comes for investment, pilots or partners. An investor looks for companies at a particular stage and in a particular industry. A corporation, technology to pilot or adopt. A public-sector representative, solutions for a region or an agency.

And each audience describes its interest in its own words:

• A startup writes: “Raise investment”↔An investor: “Find startups for the portfolio”
• A startup states its stage: MVP↔An investor is interested in Seed rounds
• A founder: “Find corporate customers and pilots”↔A corporation: “Find startups to pilot and adopt”

A person sees these connections instantly. A regular directory with filters doesn’t. So attendees don’t have to dig through thousands of profiles: the app builds each of them a personal ranking of useful contacts.

At first sign-in, each attendee answers a short questionnaire. The questions depend on their group.

• Startups give their industry, stage, revenue and what they want to get out of the summit.
• Investors give their investor type, target stages, ticket sizes, industries and goals.
• Corporations give the technologies they are looking for, the project stages they consider and the kind of collaboration they want.
• Public sector and media have questions of their own.

In total, 2,518 attendees completed the first-sign-in questionnaire. But it is only one source: the recommendation system also takes into account each person’s profile, group and professional role, how close their descriptions are in meaning, and what they do in the app. Message content, contacts and location are never used for matching.

GigaChat helps capture mutual need, not just similarity

The core difficulty is that a good business match isn’t always two similar profiles. Often the most useful people are the ones who complement each other.

We use GigaChat in the recommendation system. It plays a part in scoring compatibility and helps match answers from different audiences by meaning, recognizing supply and demand even when the two sides describe the same opportunity in completely different words.

The organizer stays in control of the logic: pairs and their strength can be adjusted or switched off.

Each matching criterion carries its own weight

For every potential pair, the system calculates an overall compatibility score, with each criterion carrying a different weight. The strongest criterion is what attendees said about their own goals in the questionnaire.

This avoids the naive “similar people first” logic. For a startup founder, a relevant investor or a corporate head of innovation ranks above another founder with a similar bio whenever they better match the attendee’s stated goal.

Roles and groups shape matching too

A shared industry isn’t enough. Two FinTech founders can genuinely help each other. But if one of them came to raise money, a venture fund partner ranks higher.

So the system takes professional roles and rules between groups into account. Some combinations get a higher priority — for example, founder → investor or founder → corporate executive. Others can be neutral or excluded from recommendations entirely. This lets the organizer shape the networking mechanics of the whole event instead of hand-picking profiles.

What a single match looks like

Picture a FinTech startup founder — call her Anna. She chose FinTech as her industry, MVP as her stage and “Raise investment” as her goal. Also in the system is Igor, a venture fund partner: industries FinTech, AI & ML, round Seed, goal “Find startups for the portfolio.” The system sees several strong signals at once — and Igor rises in Anna’s personal ranking.

GigaChat explains why this particular person could be useful

A high ranking on its own means little to an attendee. So for the top recommendations, the app generates a “Why they may be interesting” note. GigaChat receives the facts behind the recommendation and turns them into a short explanation. The model is not allowed to invent extra reasons — the text is built only from overlaps the system has actually found.

So attendees get an answer not just to “who should I look at?” but to the more important question: “why should I spend my time on this particular person?”

The next job is to keep a good match from going nowhere. The best candidates go into a personal discovery feed, where attendees mark each card as interesting or skip it. Mutual interest produces a “Business Match!” screen, with a direct path to chat or to proposing a meeting.

An invitation includes a time and place, the other attendee can accept or decline it right in the app, and a push reminder arrives before the meeting. The tech lounge ran on tables: attendees booked meetings, and lounge hosts could see how busy each zone was and seat guests at free tables.

In other words, networking is built as one continuous path:

1. understand the goal
2. find the person
3. explain the recommendation
4. get mutual interest
5. start a conversation
6. set up a meeting
7. meet at the venue

What came of the networking

A swipe in the feed is only the first step. When the interest is mutual, a business match appears and the two attendees can keep talking.

But what matters more than the swipes is whether introductions turn into real meetings. The whole path — from a swipe to a meeting at a table — is in the chart below.

Lounge hosts also seated walk-in guests at free tables 183 more times. After each meeting the app asks for a rating — and 92% of the ratings were a perfect 10.

With several thousand people on site, even simple questions quickly pile up on the support team: where is the right hall, when does a session start, who is speaking, how do I find a booth, what’s on now, where is my meeting. Most of the answers are already in the event materials — but the attendee needs them right now, not a few minutes after contacting support.

So we built a GigaChat-powered AI assistant into the app. It works from the event’s own data and answers questions about the agenda, venue, speakers and attendees.

The assistant takes questions that would otherwise go to the support team — and answers in seconds.

For all this to work, the data updates itself

Behind the user-facing features is another layer of work attendees barely notice. Attendee records sync with the summit’s registration and check-in system, so new attendees appear in the app without manual imports. The agenda — sessions, halls and speakers — refreshes from the summit website automatically, and every exhibition booth has a description, contacts and materials.

Over the life of the app, attendees started 7,743 conversations and added sessions to their personal schedules 11,201 times. But we deliberately didn’t treat these as a set of separate features: at a big event, an app only works when it connects the attendee’s whole journey.

1. understand what’s happening
2. find the right session
3. discover a useful person
4. agree to meet
5. meet
6. keep the conversation going

What we heard at the venue says more about the product than any metric.

The summit’s own app, not “yet another event app”

The app shipped under the Moscow Startup Summit brand — with its own name and icon on the App Store and Google Play, summit video on the sign-in screen, and the brand palette in light and dark themes. The home screen was built from blocks for this particular event: stories, news, introductions and the summit’s sections. It took about six weeks from kickoff to release.

All of this is platform configuration, not custom development. Organizers turn sections on and off, arrange the home screen, add their own pages and change the design from the dashboard — and the changes reach attendees without a new app release.

Finding yourself among thousands of photos

Attendees don’t have to scroll through every shot from the venue: they tap “Find your photos”, and the app finds the shots they appear in and gathers them under the “Mine” tab.

Over the two summit days, attendees spent a combined 1,855 hours in the app: swiping through introductions, arranging meetings, messaging, building personal schedules and asking the AI assistant.

The summit’s business results

This is what the summit was for: introductions were meant to turn into deals and partnerships. The official results the organizers announced on October 2:

What matters isn’t the number of features but the distance to a useful contact

For us, this project was never just about digitizing the agenda and the attendee list. The real goal was different: to shorten the distance between thousands of strangers and one specific, useful meeting.

Attendees no longer have to start with “Who among 4,500 people should I look for?” The app builds personal recommendations, and GigaChat helps find meaningful links between the goals of different audiences and explains why a particular introduction could be useful. The same product then helps turn a recommendation into a match, a match into a conversation, and a conversation into a meeting at a real table.

That is how we see the role of technology at big business events: not just showing people more information, but helping them find the people and opportunities they came for, faster.