event organizer functionaliy
attendee "icebreaker" card
standard event "concierge" functionality
mutual fit attendee list overview

SmartSummit: Event Concierge Engineering Serendipity

Inspiration

Despite the $1.022 trillion global spend on professional events in 2024 ([Custom Market Insights][1]), 80% of connections made at these gatherings dissolve before attendees even reach home. 82 % of attendees say networking is their top reason for showing up ([Amelia WordPress Booking][3]). On the supply side, B2B organisers invest 5-8 weeks of staff time per event ([Keynote Speaker and Author Jamie Turner][4]), with logistics alone consuming 20-30 hours ([EventPro][5]).

Event organizers spend 120+ hours manually matchmaking VIPs across the 1.8 million annual meetings and events in the US alone (Wonder), sponsors pour \$22.3 billion a year in North America into event partnerships ([Double the Donation][6]), yet only 35 % of marketers systematically measure sponsorship ROI ([Escalent][7]).

The problem? A fundamental disconnect between what attendees want (meaningful connections) and what events currently deliver (superficial interactions).

What it does

SmartSummit creates AI agents that truly understand each attendee (and organizers) and proactively engineer meaningful connections that last beyond the event. Unlike traditional matchmaking that produces static lists, our system:

Deep user understanding (even before signing up) through a combination of public data (LinkedIn, social profiles) and interactive conversations
Real-time opportunity detection with nudges like "Hey, someone who shares your interest in AI governance is going to this after-conference tonight—they also graduated from Stanford and play golf. Would you like to meet them?"
Facilitates meaningful introductions by evaluating bilateral fit through agent-to-agent communication
Maintains relationship continuity across multiple events with persistent memory
Provides ongoing value through post-event nurturing, multi-event interactions, and access to the global social network (though one's agent).

This approach directly addresses the 82% of attendees who rank networking as their top reason for attending events (MPIWeb) and can potentially quadruple ticket retention through AI-based intent matchmaking (Brella.io).

How we built it

We constructed an AI architecture focused on genuine understanding rather than superficial keyword matching:

Agent Epistemology Framework: A multi-dimensional knowledge model that tracks confidence in every relevant dimension, carefully separating confirmed vs. inferred knowledge about each user
Tiered Knowledge Architecture: A hybrid Postgres+Vector database with updating as new information is gathered
Temporal Intent Detection: Real-time analysis across conversations to identify matchable intents like "free for dinner tonight"
Agent-to-Agent Protocol: Structured dialogues between user agents that evaluate compatibility using a multi-factor scoring system (not just similarity)
Progressive Confidence Engine: A system that continuously improves agent understanding, filling knowledge gaps with natural conversation

The technology stack includes Postgres for structured data, vector databases for semantic similarity, Temporal for workflow orchestration, and Langgraph for agent coordination. This positions us within the rapidly growing AI in event management market, projected to reach $14.2 billion by 2033 at a CAGR of 22.9% (Market.us).

Challenges we ran into

Epistemological complexity: Creating a framework that properly distinguishes between confirmed facts vs. inferences about users without overwhelming the system
Real-time intent matching: Developing efficient algorithms to detect and match temporal intents across thousands of concurrent users
Trust calibration: Building a system that makes confident recommendations without overstepping privacy boundaries
Confidence scoring: Implementing a robust Bayesian approach to progressively update our understanding of users
Agent-to-agent communication: Designing protocols for agents to effectively evaluate bilateral fit before proposing connections

These challenges mirror the broader industry's shift toward AI adoption, with half of meeting planners already using AI, and 42 % applying it specifically to attendee matchmaking ([Cvent][8]).

Accomplishments that we're proud of

We managed to build a basic "proactive" agent in a weekend. SmartSummit has real world applications and is built with exciting technology. We integrated a proactive "filter" prompt which curates what notifications to actually send, and will automatically recommend things to the user! Using the Temporal Workflow Orchestration framework, we built a proactive recommendation service that can find all of the most relevant connections for you at any event!

What we learned

Knowledge representation matters: The difference between superficial matching and genuine understanding lies in how knowledge is structured, updated, and reasoned about
Proactive agency is key: Users respond dramatically better to natural, conversational opportunities ("Someone is free tonight...") than to static lists
Confidence metrics drive trust: By explicitly tracking confidence in each piece of information, agents can make recommendations that feel trustworthy rather than creepy
Cross-event persistence creates exponential value: The ability to maintain relationship context across multiple events creates network effects that increase with each interaction

These insights align with industry trends showing 60% of planners rating event apps as their most critical tool for operations, networking, and analytics (Nunify).