Featured engineering

Qaswar Sarfraz — Flutter & AI Mobile App Projects | Built to work,
not to demo

Production Flutter architectures, sub-second AI inference pipelines and systems written from first principles. Each one below links to its source — the previews are schematic, the code is real.

01 ✦ Flagship mobile app 2026 · shipped

Tripline — AI travel companion

Cross-platform mobile app · Flutter Web admin console

An end-to-end travel platform that generates bespoke multi-day itineraries with Groq Llama 3.3 70B, tracks multi-currency expenses and renders interactive OpenStreetMap routes at 60fps. Designed, built and shipped solo.

Sub-second AI itinerary engine: Groq Llama 3.3 70B produces day-by-day schedules under a strict JSON contract, validated before rendering.

Supabase auth & Postgres: multi-provider sign-in with row-level security policies, so a user can only ever read their own trips.

Interactive map discovery: custom OpenStreetMap styling with marker clustering and location caching.

Multi-currency budget tracker: live expense auditing with a category breakdown and an offline cache.

QR trip sharing: peer-to-peer plan sharing for collaborating travellers.

Flutter Web admin console: moderation panel for reported content and system health checks.

Flutter

Dart Groq Llama 3.3 70B

Supabase

Postgres RLS Riverpod

OpenStreetMap Flutter Web

View on GitHub ↗ See the pipeline →

02 ✦ AssemblyAI Hackathon Solo Build 2026 · Shipped & Live Demo

LectureMind — Voice-First Academic Knowledge Engine

Cross-platform Flutter App & Web · AssemblyAI Universal-3.5 Pro & Groq GPT-OSS-120B

An AI-powered voice intelligence and academic synthesis platform engineered for university scholars and graduate researchers (18+). Transforms spoken lectures, academic seminars, and document slides into deeply structured notes, vector PDFs, interactive 4-tier visual mind maps (JPG), and Socratic Bloom's taxonomy quizzes in seconds. Built solo for the Lablab.ai AssemblyAI Voice Agent Hackathon.

Real-time acoustic STT streaming: AssemblyAI Universal-3.5 Pro streaming WebSocket with domain vocabulary boosting for STEM, medicine, and Pakistani English-Urdu code-switching.

Neural STT post-processor: Groq GPT-OSS-120B strips disfluencies ("um", "matlab", "yani"), fixes misheard phonemes, and structures speech into polished academic prose with 1-tap AI term polishing.

Publication-grade structured notes & styled PDF: Deep pedagogical hierarchy (Executive Summary, Headings, Why/How breakdowns, Glossary) with instant 1-click vector PDF export and Listen Notes TTS audio recitation.

Interactive 4-tier concept mind map & instant JPG download: Subject → Pillars → Mechanisms → Details concept graph with HTML5 Canvas rasterization for dedicated 1-tap JPG download.

Socratic AI Bloom's taxonomy quiz: Generates 8–12 rigorous scenario-based analytical questions with soft Neumorphic dial gauge scoring and spaced misconception explanations.

Document & slide studio (PDF / PPT reader): Direct text extraction from lecture slide decks and research papers (.pdf, .txt, .md) to synthesize notes, mind maps, and quizzes without audio.

Flutter 3.24+

Dart Riverpod 2.6 AssemblyAI Universal-3.5 Groq GPT-OSS-120B Neumorphic Soft UI GoRouter Vector PDF Canvas JPG

Live Web Demo ↗ View on GitHub ↗

03 ✦ Independent project 2026 · in development

MentorLync — Expert Consultation & AI Mentorship

Flutter mobile app (Android & iOS) · Personal learning project

A full-stack mobile platform where Seekers book paid chat, audio, and video sessions with verified Experts — backed by real-time Supabase infrastructure, ZegoCloud calling, and a Gemini-powered AI Assistant. Built entirely solo to go deep on production-grade Flutter architecture, Riverpod state management, and backend integration.

Dual-role architecture: Seekers browse and book; Experts manage availability, earnings, and requests — all in a single app with role-gated navigation.

Real-time chat with rich media: text, images, voice notes, and file attachments over Supabase Realtime WebSocket — no user contact details ever exposed.

Background-safe audio/video calling: ZegoCloud SDK with FCM high-priority push delivers a native incoming-call UI even when the app is killed (CallKit / ConnectionService).

Gemini-powered AI Assistant: free daily quota for AI mentorship queries — visually distinct from human-expert chat and upgradeable to unlimited via Pro subscription.

Session-credit economy: purchase credit packs or subscribe to Pro for discounted pricing — all deduction logic lives in Supabase Edge Functions so the client never enforces business rules.

Safety layer: user reporting with evidence attachments, independent blocking, and Supabase RLS ensuring users only ever read their own records or those of an active-session counterparty.

Flutter

Dart Riverpod GoRouter

Supabase

Postgres RLS ZegoCloud Gemini AI FCM

View on GitHub ↗ Academic context →

ParkingManager.java

04 ✦ University capstone 2024 · completed

Smart Car Parking System

Java desktop application · NUML Islamabad

Desktop management software that automates vehicle entry, slot tracking and receipt generation across multiple floors. This is the project where object-oriented design stopped being theory.

Live parking grid: real-time visual slot availability across floors.

Automated entry & exit: sequential ticket generation with vehicle categorisation.

Billing engine: hourly tariff calculation with itemised receipt printing.

Persistence: file-backed storage that survives a restart with no data loss.

Java Swing GUI OOP patterns File I/O

View on GitHub ↗ Academic context →

On the bench

Currently building

Document AI Assistant

Mobile-first document chat with retrieval

In development

A Flutter client over a Python service that parses documents, embeds them, and answers questions against the retrieved context. The honest status: the retrieval layer works, the mobile UX is still being built, and there is nothing to demo yet.

Flutter