Home/Case Studies/Flash
CASE STUDYE-COMMERCE · EMAIL TO ENTITY EXTRACTION PIPELINE

Flash

A privacy-first shopping platform that turns inbox emails into organised orders, tracking purchases across e-commerce, food, and travel while rewarding users with gamified loyalty.

147Weeks operated
10k+Hours of work
12Engineers

The brief

Build Flash into a comprehensive, privacy-first shopping intelligence platform that consolidates all purchase data from users' inboxes into a single organised view, delivers actionable spending insights, and rewards users with a loyalty/deals engine — differentiating on a dedicated spam-free email address and universal order tracking across all merchants and travel categories.

What we built

A full-stack consumer shopping intelligence platform spanning:

  • a dedicated private email address with Gmail-integration pipeline that auto-ingests and classifies transactional emails into typed order entities (e-commerce, food, travel, ride, hotel)
  • a universal order tracking and spending insights dashboard with bar/pie/line charts
  • a gamified loyalty engine with streaks, challenges, milestone campaigns, cashback, Flash Drop surprise boxes, and a Flash Plus auto-redemption subscription tier
  • a self-serve deal/voucher catalogue with admin portal (Retool/AppSmith) for merchant coupon and creative management
  • a cross-platform notifications and event-log infrastructure
  • a Chrome browser extension with PDP overlay pill, side-panel AI chat, and cross-sell intelligence widgets (ingredients, skin concern, suitability, specs-decoded)
  • a React/Next.js web app with Core Web Vitals optimisation; and
  • TWID payment platform integration for reward-point checkout redemption. The system serves both Indian and US markets with full i18n, country-aware APIs, and dual-currency support.
⊛ Live in production

Full Flutter mobile app (iOS + Android) with inbox, order tracking, insights, rewards, gamification, and onboarding live on prod; Java microservices (email pipeline, entity extraction, categorisation, engagement/campaign engine, comms/PN service) on AWS ECS; self-serve admin portal on Retool/AppSmith; Chrome extension with PDP pill and AI cross-sell widgets; React/Next.js web app with optimised Core Web Vitals; TWID redemption integration and Flash Plus auto-redemption subscription tier deployed to prod.

How it was built, phase by phase.

8 workstreams across 147 weeks of operated delivery

  1. (001)BUILDWeek 1–60

    Gmail Integration & Email Processing Pipeline

    Multi-service pipeline ingesting Gmail data via OAuth consent, watch subscriptions, SQS queues, and Lambda functions to extract, categorize, and classify transactional emails in near-real-time.

    ✓ Full async email ingestion pipeline live on prod; Gmail watch renewals, merchant classification, type extraction.

    JavaAWS SQSAWS SNSAWS LambdaAWS ECSRedis
  2. (002)BUILDWeek 1–52

    Order Entity Extraction & Universal Order Tracking

    Extraction and structuring of order entities (e-commerce, food, travel, ride) from emails into typed entities with tracking status, delivery milestones, and rich metadata exposed via API.

    ✓ End-to-end entity extraction for shopping, food.

    JavaPython-DjangoAWS SQSPostgreSQLRedisJOOQ
  3. (003)BUILDWeek 1–80

    Flash Inbox — Dedicated Shopping Email Client

    Full-featured mobile inbox for Flash's proprietary email address, including compose, reply, forward, threading, search, attachments, trash, spam management, and outbox with retry logic.

    ✓ Fully functional dedicated Flash email client with batcher-based outbox retry, attachment caching, threading, and search live in production.

    FlutterDartHiveFirebaseAWS SQSJava
  4. (004)BUILDWeek 1–130

    Rewards, Gamification & Campaign Engine

    Multi-tier loyalty system comprising streaks, challenges, milestone campaigns, cashback campaigns, surprise boxes (Flash Drop), coin ledger, redemption flows, and a rule engine for priority-based campaign execution.

    ✓ Full campaign engine live with priority-based rule execution, milestone/challenge/cashback campaigns, Flash Drop, Flash Plus auto-redemption.

    JavaAWS SQSRedisPostgreSQLRetoolApp-Smith
  5. (005)BUILDWeek 1–90

    Focus Widget System & Home Feed Personalisation

    Dynamic card-based home feed with dismissible animated focus widgets, remote-config-driven rendering, shimmer states, undo snackbar, and per-widget analytics tracking impressions and actions.

    ✓ Animated, dismissible focus widget carousel live with remote-config rendering, per-widget analytics, shimmer loading, and undo-snackbar UX.

    FlutterDartFirebase Remote ConfigHivePercept
  6. (006)BUILDWeek 8–60

    Spending Insights & Analytics Dashboard

    In-app analytics showing spending breakdown by brand/category, bar/pie/line charts with timeline filters, spam metrics, top brands, rank widget, and multi-dimensional insights for user financial awareness.

    ✓ Full insights tab live with bar/pie/line charts, brand breakdown, spam metrics, timeline filters, and Mixpanel/Percept analytics instrumentation.

    FlutterDartFirebaseMixpanelMoEngagePercept
  7. (007)BUILDWeek 40–100

    Deals & Rewards Catalogue (Self-Serve Portal)

    Merchant deal catalogue with variant management, coupon upload (bulk CSV + single), creative S3 upload, tag-based filtering, country-aware visibility, and admin portal built on Retool/AppSmith for ops team.

    ✓ Full catalogue management portal live on Retool with deal creation, variant/coupon/creative upload, tag filters, country-aware listing.

    JavaRetoolApp-SmithAWS S3PostgreSQLPython-Django
  8. (008)INTEGRATEWeek 40–70

    Internationalisation & Multi-Country Support

    Full i18n implementation covering locale detection, country code selection, currency formatting, timezone-aware API headers, country-aware deal/package/reward visibility, and TNC/privacy policy routing by country.

    ✓ Full IN/US dual-market support live with locale-aware headers, country-specific TNC/privacy, currency formatting.

    FlutterDartFirebase Remote ConfigJavaPython-Django

Stop renting hours. Start running functions.

Pick the function you want off your plate. We map the brain and name the outcome before you commit.