Outrizz AI: Event Engagement Platform

Outrizz AI runs one pipeline in two phases. Before the event, agents source, qualify and talk to prospects to drive attendance. During it, the booth team captures visitors on a phone and follow-up goes out within hours. Built together with the Outrizz team.

Sourcing

The pipeline collects from LinkedIn, industry databases, company sites and other public sources, then enriches each contact with company data and verifies it. Records refresh continuously, because half the value of an event list is that it is still correct on the day.

Qualification runs on top: an LLM scores fit against the client's ideal customer profile, reads behaviour for buying signals and attendance probability, and ranks the list. Clients tune the scoring on their own historical conversions.

Outreach agents

Messages are written per prospect from their profile, their company, and why this event is relevant to them. Timing is coordinated across email and LinkedIn so the same person is not hit on both channels at once. Agents handle common objections in the thread and book the meeting when the prospect agrees.

Capture at the booth

The mobile app takes a photo of a card and pulls out the contact fields, or takes voice notes, or plain manual entry when the light is bad. Each new record is enriched and checked against existing prospects so a lead sourced two weeks ago does not become a second contact.

Follow-up references the actual booth conversation, so the first message after the event reads like a continuation rather than a blast.

Training agents on the client's data

Every client gets a sandbox. They train agents on past conversions, winning sales calls and their own messaging, A/B test agent configurations against each other, and watch engagement and conversion per variant. The brand voice and the sales playbook stay consistent across everything the agents send.

Different jobs get different models. Sourcing, qualification, conversation and follow-up each route to the model that does that job best, with failover to another provider when one is unavailable.

Architecture

A Turbo monorepo holds the web app, the mobile app and shared packages for AI agents, database, UI and validation, so business logic is written once. tRPC generates the client from the server, Drizzle generates types from the schema, and Zod validates every payload at runtime, including model output.

Data sits in Supabase Postgres. Mobile is React Native on Expo, with over-the-air updates, camera and audio access, and local persistence so the app keeps working when the venue wifi does not. Web is Next.js, installable as a PWA.

Campaigns run across multiple events and venues, every lead carries its event attribution, and teams work in the same account with role-based permissions. Data syncs both ways with the client's CRM, with the full interaction history attached.

Stack

Next.js and React Native on the front, Supabase and Drizzle for data, tRPC between them, TypeScript strict everywhere. On the AI side: several LLM providers behind one interface for conversation and tool calling, a vision model for card OCR, and multi-language speech transcription for booth audio.