ayaan = {
"role": "AI Systems Engineer + Full-Stack + Co-Founder of ForshaLabs",
"location": "Mumbai, India 🇮🇳",
"education": "B.E. Computer Engineering — Rizvi College (GPA 9.09/10)",
"building": "MobCloudX — AI QoE Intelligence for Adaptive Video Streaming",
"research": "ACM MMSys / IEEE IWQoS paper (under submission)",
"won": ["3× National Hackathon Winner", "IBM SkillsBuild Top 30", "Techathon Runner-Up"],
"superpower": "End-to-end solo systems that span ML → Infra → Cryptography"
}I don't just integrate AI — I build the perception layer, the adaptation layer, and the verification layer that makes AI trustworthy in production.
🎬 MobCloudX — AI QoE Intelligence Platform
The most technically complex solo project I've shipped
An end-to-end platform for adaptive video streaming that monitors, adapts, and cryptographically proves quality in real time.
┌─ Android SDK ──────────────────────────────────┐
│ CNN-LSTM frame scoring │ PPO ABR Agent │
│ RealESRGAN 4× SR (12ms) │ ZK Proof Client │
└──────────────┬─────────────────────────────────┘
│ Apache Kafka
┌──────────────▼─────────────────────────────────┐
│ FastAPI Inference │ FedAvg FL Server (247 clients) │
│ XGBoost Predictor │ Groth16 ZK Circuit (Circom) │
│ MongoDB Atlas │ Polygon Goerli Blockchain │
└──────────────────────────────────────────────────────┘
| Metric | Result |
|---|---|
| 🎯 QoE Composite Score | 94.2 / 100 |
| 📺 VMAF Perceptual Score | 93.9 / 100 |
| ⚡ RealESRGAN SR improvement | 61.2 → 87.3 VMAF (+42.6%) |
| 🔮 LSTM Bandwidth Prediction | 93.7% accuracy |
| 🔒 FL Communication Overhead | ↓89% vs centralised |
| ⛓️ ZK Proof Mode | Groth16 · verified true · Polygon anchored |
TypeScript Python Kotlin FastAPI Kafka MongoDB Docker AWS PyTorch ONNX Circom
Languages
AI / ML / Computer Vision
Backend + Distributed Systems
DevOps + Cloud + Observability
Frontend + Mobile
|
AI QoE Intelligence Platform End-to-end adaptive video streaming platform with CNN-LSTM quality scoring, PPO RL adaptation agent, RealESRGAN 4× super-resolution, FedAvg federated learning with DP, and Groth16 ZK-proof SLA verification on Polygon.
|
Hindi/Hinglish Voice Ledger — Hackathon Build 24-hour hackathon MVP for street vendor financial tracking. LangGraph agentic pipeline (transcribe→guardrail→extract→anomaly→memory), Groq STT, ElevenLabs voice agent (Arjun persona), multilingual-e5-small embeddings, Qdrant vector store.
|
Smart Crop Health Monitoring
YOLOv11-based pest detection across 22 classes with 80% accuracy, pruning + quantization for 65% faster inference, Random Forest soil nutrient profiling at 95% accuracy, and ONNX-accelerated deployment.
YOLOv11 ONNX Random Forest Docker CI/CD
| 🏅 Achievement | Details |
|---|---|
| 🥈 Hackathon Runner-Up | Vaultix — Secure Cloud Storage · Techathon 2025 |
| 🏆 Top 30 National | IBM SkillsBuild Maharashtra Hackathon 2025 |
| 🥇 3× Hackathon Winner | National-level competitive builds |
| 💻 Competitive Programming | 250+ LeetCode · 300+ GFG/Code360 |
| 🏅 DSA Finalist | TechGig Competitive Programming Event |
| 🎓 Academic Excellence | GPA 9.09/10 — Computer Engineering |
| | 📄 Research Paper | MobCloudX — ACM MMSys / IEEE IWQoS (under submission) |
▸ GPU Kernel Optimization → Triton · CUDA · Flash Attention internals
▸ Inference Serving → TensorRT · NVIDIA Triton Inference Server
▸ Kubernetes + GPU Orch. → MIG · DCGM · GPU-aware scheduling
▸ Distributed Systems → MIT 6.824 · Raft consensus · CRDTs
▸ ZK Cryptography → Groth16 · Plonk · zkML systems
Check my Hashnode for system design write-ups, ML deployment guides, and lessons from building production AI systems solo.
I'm open to AI/ML engineering roles, distributed systems positions, and research collaborations in video intelligence, federated learning, or GPU inference infrastructure.
Email: amjad.quasmi@gmail.com | Mumbai, India | Open to remote