MU
MUHAMMAD USMANSOFTWARE ENGINEER • FLUTTER & AI
BACK TO ALL PROJECTS
Think Study Learn Ltd.Flutter & Backend Developer2026 – Present

Think Study Learn (TSL)

AI Educational Super-App for 50,000+ Active Students

Active Installs50,000+
Quiz Accuracy99.4%
OCR Speed<1.2s
App Rating4.8 ★
THE ENGINEERING CHALLENGE

Students needed instant answers and contextual explanations for complex math, science, and history questions from photo uploads without experiencing token timeout errors or hallucinated responses.

THE ARCHITECTURAL SOLUTION

Architected a Retrieval-Augmented Generation (RAG) pipeline connecting AWS Bedrock Claude models with Google ML Kit OCR scanning on the client device. Reduced LLM token costs by 40% through localized document pre-processing.

SYSTEM ARCHITECTURE & DATA FLOW

Edge OCR & AWS Bedrock RAG Engine

Client-side text recognition coupled with cloud vector indexing and multi-tier prompt caching.

01 // STAGEGoogle ML Kit Edge Text Recognition (iOS & Android)
02 // STAGENode.js Microservices & Token Rate Limiter
03 // STAGEAWS Bedrock Claude 3.5 Sonnet Integration
04 // STAGEVectorized Knowledge Base & Embeddings
05 // STAGEInteractive Quiz Engine & Progress Tracking
KEY ENGINEERING HIGHLIGHTS & SOLVED BOTTLENECKS

Sub-Second OCR & Math Parsing

Integrated on-device ML Kit text recognition with LaTeX formatting for crystal-clear mathematical equation rendering.

Google ML Kit • LaTeX Flutter • Dart

AWS Bedrock AI Tutor with Streaming Responses

Implemented Server-Sent Events (SSE) in Flutter to render typing tutor responses with zero perceived latency.

AWS Bedrock • SSE Streams • Node.js

Automated Google Play / App Store Deployments

Configured GitHub Actions CI/CD pipelines producing signed AAB and IPA bundles automatically on version tag pushes.

Fastlane • GitHub Actions • Firebase App Distribution
TECHNOLOGY STACK & INFRASTRUCTURE
FlutterAWS BedrockNode.jsGoogle ML KitFirebase FirestoreGitHub ActionsIn-App Purchases