Case study · QuickBiteA creator-led food discovery app where reels turn into verified visits

Creators film real food videos, people claim offers, restaurants pay for the visits a reel drives — a social app on top of a backend built like a payments system.

Built at GravitonesIn development

Industry: Food & social
Location: Bhubaneswar, India
Timeline: Sep 2026 – present
Year: 2026

screens in the React Native app (Expo Router): 0+; screens in the React Native app (Expo Router)
admin panel screens for moderation, catalogue and payouts: 0+; admin panel screens for moderation, catalogue and payouts
backend modules in one typed, observable API: 0; backend modules in one typed, observable API
Indic scripts covered by phonetic search: 0; Indic scripts covered by phonetic search

The challenge

What the client needed.

QuickBite connects three groups around one loop: creators film real videos at real restaurants, people watch, save and claim offers, and restaurants pay creators for the footfall they drive. The loop only works if a visit can be proven — and because money flows from restaurants through the platform to creators, the backend has to be built like a payments system from day one while the app feels like social media.

My role

Lead developer — product architecture, backend, React Native app, admin panel and landing site

The approach

How I built it.

01
Designed the backend as a typed modular monolith — user, post, restaurant, analytics and admin modules in one deployable — with strict boundaries so any module can be extracted later.
02
Kept PostgreSQL with PostGIS as the system of record and the only home of money, and served feeds, counters and inboxes from MongoDB read models fed by an outbox.
03
Closed the loop with QR offer redemption: the in-store scan is the event that proves footfall and drives creator earnings, guarded by velocity limits.
04
Built a React Native app with Expo Router — reels, stories, offers, saves, creator studio, campaigns and payouts — plus a React admin panel and a Next.js landing site.
05
Engineered trust and quality in: phonetic search across nine Indic scripts, perceptual hashing of removed media, ring and spam detection, push quiet hours, OpenTelemetry tracing and Prometheus metrics.

Key engineering decisions

The hard problems, and how I solved them.

The decisions that made the difference — each one a trade-off chosen on purpose.

01
Proving a reel caused a visit
Problem
Restaurants can't tell which creator content actually brings people through the door.
Decision
Offers are redeemed by an in-store QR scan and attributed to the creator whose reel produced the claim.
Outcome
Creators can be paid per verified visit rather than per post.
02
Feeds that never touch the ledger
Problem
Social read volume would overwhelm the database that holds money.
Decision
Postgres for writes and money; MongoDB read models updated through an outbox; cursor pagination on every feed.
Outcome
Scrolling scales independently of the payments core.
03
Search that speaks Indian languages
Problem
People search for food in Odia, Hindi and transliterated English — exact-match search fails them.
Decision
Phonetic keys over nine Indic scripts, a synonym table mined from search logs and city-ranked typeahead on Postgres.
Outcome
Relevant results without paying for a search cluster before scale demands it.
04
Integrity before launch
Problem
Reward-driven apps attract fake accounts, like rings and re-uploaded content.
Decision
Perceptual hashes of removed media, per-device velocity limits, ring detection and image moderation with AWS Rekognition.
Outcome
Abuse controls exist before the incentives that invite abuse go live.

What's built

What's built so far.

Mobile app
iOS & Android: reels and stories, restaurant pages, offers and QR redemption, saves, DMs and creator studio.
Backend API
TypeScript modular monolith on PostgreSQL + PostGIS, MongoDB read models, Redis and WebSockets.
Admin panel
Moderation, upgrades for creators and restaurants, catalogue, search synonyms, campaigns and payouts.
Landing site
Next.js marketing site with legal, cookie and account-deletion pages.
Media pipeline
Direct-to-S3 uploads, on-device video processing and automated moderation.
Observability
OpenTelemetry traces, Prometheus metrics, alerts and structured logs.