MeeeetUp FaceID — Face Recognition Check-In Platform

A face-ID platform for events, offices, hotels, and venues — people register their face once through an invite link, then cameras at the entrance recognise them and mark them as arrived. No tickets, no scanning, no app to install. Currently in active development.

Overview

FaceID is sold as a provider service: other products integrate with it instead of building face recognition themselves. An admin creates a project, defines a scope (an event, a stay, an office visit), and shares a registration link. Registered people are then identified automatically at any camera paired to that scope, while unrecognised faces are still recorded so operators can see everyone who walked in.

The platform ships as three parts: the multi-tenant SaaS backend and consoles, a distributable face-capture SDK, and first-party capture clients for browser, Android, and desktop.

Key Achievements

• Cross-platform capture SDK — One shared detection and tracking pipeline published as three packages (platform-agnostic core, web, React Native), so a new client platform reuses the same capture behaviour instead of reimplementing it
• On-device frame selection — Faces are tracked locally and scored for how frontal they are; only the best frame per person per 10-second window is uploaded, which cuts recognition traffic and raises match quality versus sending every frame
• Transport-agnostic SDK design — The SDK emits capture buffers and lets each consumer own its network layer, which let the platform change its ingest model without a single SDK rewrite for existing integrators
• Self-serve registration — People enrol themselves through a link, removing manual guest-list data entry from operators
• Auto-generated integrator docs — The public API serves its own OpenAPI document and an llms.txt guide generated from the route definitions, so docs can never drift from the shipped API
• Bilingual product — Japanese-first UI with English fallback across every console

Key Features

• Automatic check-in — Cameras recognise registered people and mark attendance with no interaction at the door
• Multi-tenant workspaces — Organisations own projects, scopes, cameras, and their own isolated face collections
• Scope-based registration forms — Each scope defines its own form, giving one project separate guest lists per event or visit
• Unknown-face capture — Unrecognised visitors are recorded as provisional identities rather than dropped
• Fine-grained permissions — Per-action permission catalog for members and machine credentials instead of coarse role names
• External API — Versioned public API with scoped machine credentials for partner products
• Capture clients — Browser PWA, React Native Android client with a native frame processor, and a desktop client
• Admin console — Project, scope, camera, credential, and people management with server-side searchable, sortable tables and attendance reporting

Tech Stack

Frontend: React, Vite, TanStack Router, TanStack Query, Next.js (marketing), Tailwind CSS, shadcn, i18next

Backend: Hono, Bun, TypeScript, Zod, Drizzle ORM, PostgreSQL, layered repository / domain / service architecture

Face Recognition: AWS Rekognition (server-side identification), MediaPipe BlazeFace (web detection), ML Kit (native detection)

Mobile: React Native, Vision Camera with a native frame processor

Tooling: Turborepo, pnpm, Docker, OpenAPI

Role

Full-Stack Engineer — Owned the API and data layer, the admin and registration consoles, the public versioned API, the face-capture SDK, and the first-party capture clients.