name: Aashir Athar
role: Senior React Native Engineer 📱
side_quest: Open-Source Builder · AI Tinkerer · UI/UX Practitioner 🎨
location: Lahore, Pakistan 🇵🇰
experience: 4+ years shipping production mobile apps to 50,000+ users
day_job: Expo SDK 55 · New Architecture · TypeScript strict · Offline-first
night_job: On-device AI · Native Kotlin modules · Self-training LLMs
fuel: ☕ × engineering judgment²
philosophy: "AI-multiplied output, with engineering judgment still in the driver's seat."📱 I ship mobile apps that don't crash, don't drop frames, and don't ask for your trust without earning it. Three years deep in React Native, with production work serving 50K+ users across the United States and Pakistan. I author native Kotlin modules, publish them to NPM, run TensorFlow Lite on-device, and tune for 120 FPS on 2 GB-RAM Android phones — because good enough on a flagship isn't good enough for the users who actually need the app to work. Engineered with care, validated with telemetry. 💙
📱 Mobile (the primary focus)
💻 Languages, State & Data
☁️ Backend & Infrastructure
🤖 AI / ML Playground
🎨 Design, Media & DevOps
💡 Want the snake animation? Add the GitHub Action
Platane/snk— workflow file template is at the bottom of this README.
⭐ Main Project — xMind
"A social network engineered like a system, not a feed." — A production-grade open-source social platform built end-to-end.
|
A reference-implementation social platform — Expo client, stateless Node.js / Express API on Vercel, MongoDB Atlas, and a 12-layer on-device feed ranking engine combining TF-IDF cosine similarity, engagement velocity, exponential temporal decay, social-graph signals, and MMR reranking — sustaining sub-3ms per-page ranking latency. Token-driven design system (primitive → semantic → component) with platform-aware translucency: iOS 26 Liquid Glass, BlurView fallback, flat tinted surfaces on Android — single-point theme switching across the entire application. Cold start <1.5s on Pixel 5. |
|
Real-time blood donor matching with expanding geo-ring escalation (1→5→15→30→50km→nationwide) via PostGIS |
AI-powered fintech that builds itself — Pakistani bank/wallet SMS (HBL, Meezan, EasyPaisa, JazzCash + 22 more) parsed on-device into a private ledger. Zero cloud, zero analytics, all processing local. Trilingual (EN/UR/HI with RTL). # On-device parsing + Groq AI assistant + 3-layer rate limiter |
|
Published Expo native module on NPM. Converts raw banking SMS into typed transaction objects across 60+ financial institutions (PK, IN, BD, GCC). 9-factor weighted confidence model, ref-counted listener pattern, dependency-free TS parser. import { addTransactionListener } from 'expo-transaction-sms-reader' |
Conservation app for WWF Pakistan rangers in connectivity-dead regions. On-device MobileNetV2 leaf-health classifier with HSV fallback. Closed active-learning loop → Python retraining → OTA model updates via Supabase Storage. PostGIS polygon incursion detection. |
|
A genuinely self-improving LLM built from scratch in TypeScript — not a vendor wrapper. DeepSeek-V3-class: MoE top-K routing, MLA (~10× KV cache reduction), GQA, RoPE, RMSNorm, SwiGLU. Continual learning every 5 min with eval-gated commits + LoRA rollback. |
🚗 fuelio
Vehicle management — fuel consumption, efficiency calculations, service history tracking. Entirely on-device. No cloud, no analytics, no telemetry. Built for users who want to own their data without compromise. |
🎧 Crate
Offline-first music player for Creative Commons & public-domain music (Jamendo · Internet Archive · Audius). Paste a link, download legally at best quality, play fully offline with background + lock-screen controls. License-gated by design. |
Autonomous crypto signal engine on free GitHub Actions cron — polls Kraken, runs a self-learning ML ensemble (Bayes · kNN · logistic · Q-learning) on BTC/ETH, ships Telegram alerts. Honest 52–62% out-of-sample accuracy, pure stdlib. |
+ 🧠 xMind — scaling on-device ranking, multi-cluster Mongo, real-time presence (main focus)
+ 📦 expo-transaction-sms-reader v2 — iOS support, more regions, plugin parser API
+ 🤖 MindeesAI — closing the self-training loop, GRPO benchmarks, multi-provider failover
+ 🌍 Cross-platform New Architecture migration patterns for legacy Expo SDK ≤49 apps
! 🎨 Liquid Glass design language on iOS 26 + faithful Android Material You parity
- 🔐 Privacy-first mobile: SMS, biometrics, secure enclaves, zero analytics SDKs ┌──────────────┐ ┌──────────────┐ ┌──────────────┐
│ PIXELS & │ ───▶ │ LOGIC & │ ───▶ │ ON-DEVICE │
│ FIGMA │ │ TYPE-SAFETY │ │ INTELLIGENCE│
└──────────────┘ └──────────────┘ └──────────────┘
🎨 💻 🤖
"If you can paint "If you can model "If you can train
a frame in Figma, a backend in TS, a model in Python,
you can ship UI." you can ship apps." you can run it offline."
Mobile engineering doesn't stop at JavaScript. The same instincts that move pixels in Figma — understand the user, eliminate friction, iterate, measure — are the ones I bring to every API call, every list cell, every animation worklet. That's how an app earns 50,000+ users and a clean post-launch crash record. 📱✨
Every AI tool earns its place. None of them replace engineering judgment.
A purpose-built environment — tailored Claude agents, skill packs, prompt libraries, and UI/UX best-practice modules — that compounds team velocity without trading away code quality. This isn't AI-assisted work; it's AI-multiplied output with engineering judgment still in the driver's seat.
-000000?style=for-the-badge&logo=x&logoColor=white){kind=icon}
Open to senior React Native roles, mobile architecture consulting, open-source collaborations on Expo modules, AI-on-device features, or any production-grade mobile build that demands engineering judgment over framework juggling.
📱 Crafted with engineering judgment, shipped with EAS, debugged with ☕ at 2 AM. — © Aashir Athar