MediaIndo-UK6 months

Event Photo Face-Match App - Indo-UK

How a selfie-to-photos web app helped guests at 10,000+ events find their own photos in seconds, with self-hosted face recognition and privacy built in.

Indo-UK event photography and media company (name withheld)November 2026Consulting CTO

A wedding guest taking a selfie to find their own photos in an event gallery

Overview

After a wedding, conference or college fest, guests had to scroll through hundreds of photos to find the few they appear in. As consulting CTO for an Indo-UK event-media company, I designed a web app where guests take one selfie and instantly get every photo they appear in, with no app and no sign-in.

Technologies Used

Ollama

Self-hosted face-recognition model, run offline

MongoDB Atlas Vector Search

Face embeddings and similarity search

MongoDB

Events, photo records and match results

Node.js

Events, uploads and search API

React.js

Organiser and guest web app

AWS S3

Mid-resolution photo storage

The Challenge

After a wedding, conference or college fest, guests had to scroll through hundreds or thousands of photos to find the few they appear in.

Galleries are huge. A single big Indian wedding can produce thousands of photos across several days and functions.
Guests give up. Most only want their own photos, and few will scroll the full gallery to find them.
Organisers carry the load. Photographers and planners field requests like "please send my photos" for weeks after the event.
Sharing is messy. Photos travel through generic drive links and messaging groups, with no control over who sees what.

The client wanted guests to find their own photos in seconds, without installing an app or creating an account.

A guest overwhelmed by a gallery of thousands of event photos

My Role

As consulting CTO, I owned the technical design and guided a team of 5+ engineers from idea to launch over 6 months.

Architecture: the React web app, Node.js backend, MongoDB data model, the offline face-recognition pipeline and vector search.
Face-matching design: how event photos are indexed, how a guest's selfie is matched, and how results stay limited to one event.
Privacy: consent, how face data is stored, who can see which photos, and when data is deleted.
Delivery oversight: guiding the development team, reviewing code and planning the launch.

Architecture

The app has two simple journeys, one for the organiser and one for the guest, joined by face embeddings that belong to a single event.

Face-match flow: organiser uploads photos, a guest's selfie is matched within one event

Create the event
The organiser creates an event in the React web app and gets a unique link to share with guests.
Upload and index
Event photos are converted to mid resolution to save storage, then uploaded to AWS S3. A face-recognition model, served offline with Ollama on our own servers, detects each face and turns it into a face embedding: a list of numbers that describes the face. The embeddings are stored in MongoDB Atlas Vector Search, tagged with the event.
Guest opens the link
Guests open the shared link on their phone, with no app to install and no sign-in.
One selfie
After giving consent on screen, the guest takes a selfie. The same model turns it into an embedding, and a vector search finds the closest matching faces within that event only.
Instant results
The guest sees every photo they appear in, ready to download or share.

Because the expensive work (creating face embeddings) happens once per photo at upload, each guest search stays fast and cheap.

Privacy by Design

A face is biometric data, so the app was built to match guests within one event and nothing more.

Matching stays inside the event. A guest's selfie is compared only with that event's photos, never across events or clients. Recognition runs offline on our own servers, so photos and face data are never sent to a third-party AI service.
The selfie is the key. Guests don't sign in: their own face is what unlocks their photos. If a guest's face isn't in the event's photos, the app shows nothing, so nobody can browse a gallery they don't appear in. Guests give their consent on screen before the selfie is taken.
Access by link. Each event has its own unique link, shared by the organiser.
Clear retention. All selfies and galleries are deleted after about 3 months, or at the end of the organiser's plan.

A consent screen before a guest's selfie is used to find event photos

The same feature is regulated almost everywhere, so these choices carry across markets:

Region	How face data is treated	What it means for the design
Europe and UK	GDPR: biometric data for identification is special-category data.	Explicit consent before the selfie; EU or UK hosting; deletion on request.
Middle East	UAE and Saudi PDPL: biometric data is sensitive data.	Clear consent and a stated purpose; check data-residency rules in Saudi Arabia.
Africa	POPIA: biometric information is special personal information.	Consent and a lawful purpose; strict limits on reuse.
India	DPDP Act: personal data processed with consent for a stated purpose.	Consent screen, purpose shown up front, deletion when the purpose ends.
South East Asia	Singapore, Thailand, Vietnam, Malaysia and others protect biometric data.	Consent, local hosting where required, short retention.

Trade-offs We Made

Decision	What we gained	What we gave up
A self-hosted model served with Ollama, running offline	No per-image or per-search fees; face data never leaves our infrastructure.	More engineering to set up, tune accuracy and scale processing.
Face embeddings in MongoDB Atlas Vector Search	Fast similarity search next to the event data already in MongoDB.	Vector-index tuning as galleries grow.
A web link and a selfie, with no app and no sign-in	Guests find photos in seconds, on any phone, and only ever see photos they appear in.	No app-store presence or push notifications.
Matching one event at a time	Simpler privacy story and faster searches.	No "all my photos across events" view for repeat guests.
Mid-resolution photo copies	Lower storage costs across millions of images.	Guests get mid-resolution photos rather than full-size originals.

What we deliberately did not do: send guests' photos and selfies to a third-party face-recognition API. Running it ourselves kept costs flat as galleries grew and kept face data under our control.

Results & Impact

Across 10,000+ events, guests found their own photos by taking one selfie, instead of scrolling galleries of 500 to 800 photos.

Events hosted

10K+

Each with 500 to 800 photos

Photos processed

5M+

Calculated from 10K+ events at 500 to 800 photos each

Guest selfie searches

1M+

Calculated from 100 to 200+ selfies per event

Data retention

~3 months

Or as per the organiser's plan

Business Impact

10,000+ events hosted on the platform

5M+ event photos indexed with self-hosted face recognition

1M+ guest selfies matched within their own event

No app and no sign-in: the selfie is the key

Photos and face data never sent to a third-party AI service

Selfies and galleries deleted automatically after about 3 months

Fewer "please send my photos" requests for organisers

Where This Pattern Applies

The same selfie-to-photos flow works for any event with a large crowd and a professional photographer, in any market:

Weddings and family functions, from multi-day Indian weddings to Gulf and African celebrations.
Conferences and trade shows, such as tech and startup events in Dubai, Lisbon or Singapore.
Marathons, sports days and school or college events.
Corporate offsites, product launches and award nights.

Key Takeaways

A self-hosted face model with vector search can beat per-image cloud fees, and keeps face data on your own infrastructure.

Keep matching inside one event, and treat every selfie as biometric data with consent and deletion built in.

Let guests use a link and a selfie, with no app to install and no sign-in.

Do the heavy work once at upload, so every guest search stays fast and cheap.

Mid-resolution copies and automatic deletion keep storage costs and data risk down.

Planning an AI Feature That Handles Personal Data?

With 26+ years of building platforms, I help founders design AI features that handle faces, voices or other personal data to be useful, affordable and compliant in every market they serve.

Book Free Consultation View More Case Studies

Illustration of a startup product team choosing one useful AI feature from many flashy options

TechnicalOctober 22, 20267 min read

Where AI Actually Earns Its Place in an MVP

Read Article

Related Case Studies

Illustration of photographers and clients across Europe connected through a creative services marketplace

MarketplaceEurope

Creative Services Marketplace - Europe

How we architected a marketplace where photographers, videographers and creative studios across Europe find work, deliver galleries and get paid safely.

Illustration of EV chargers from different makers connected to one white-label charging management platform

CleanTechIndia

White-Label EV Charging Platform - India

How we architected a vendor-neutral, white-label charging management platform that lets any EV charging operator launch under its own brand.

Illustration of a gaming company and a freelancer transacting safely through an escrow vault

iGamingAfrica

Gaming Freelancer Marketplace - Africa

How we architected RI Experts, a South African marketplace where iGaming companies hire specialist freelancers worldwide, with escrow on every job.

Illustration of families connecting through a private community matrimony app on their phones

MatrimonyIndia

Community Matrimony Platform - India

How we built and launched Rawa Rajput Matrimony, a trusted matchmaking PWA for one community, in a form any family can open from a WhatsApp link.

Short-video creator app with live streaming and virtual gifting on a smartphone

MediaIndia

Short-Video Streaming App - India

How a short-video, live-streaming and coin-gifting app for Indian performers reached 1M+ downloads, with smooth playback on mobile networks and payouts to creators.