I'm a Computer Science student at UET Lahore (graduating June 2026) who builds at the intersection of AI/ML and backend engineering. I architect multimodal deep learning systems, production-grade REST APIs, and microservice backends — not just toy demos.

• 🧠 Currently working on: Multimodal Emotion Recognition — a 5-layer production AI pipeline (Flutter → FastAPI microservices → fusion gateway → therapeutic LLM with crisis protocols)
• ⚙️ Also building clean .NET backends with ASP.NET Core, EF Core, and JWT auth
• 🎓 Served as Undergraduate Lab Assistant at UET, teaching C++ to first-year CS students
• 📍 Based in Lahore, Pakistan

TensorFlow FastAPI Microservices Flutter NLP Computer Vision

A safety-critical AI system for multimodal emotion detection across audio, visual, and text streams.

• Architected a 5-layer production pipeline: Flutter mobile client → 3 FastAPI microservices (text/audio/vision) → fusion gateway with weighted softmax + contradiction detection
• Engineered dual-layer safety screening: regex pre-gates + 8B LLM classifier routing to a 70B therapeutic LLM with crisis protocol and locale-aware helpline injection
• Optimized CPU inference via TFLite float16 quantization (2.3× latency reduction), @tf.function graph compilation, and zero-disk FFmpeg RAM pipelines
• Achieved 100% masked-distress detection in multimodal vs. 0% text-only (validated via staged vignette replay, n=30, with clinician-informed oracle labels)

ASP.NET Core JWT EF Core SQL Server

Production-grade auth service with user registration, secure password hashing, JWT authorization, and Swagger documentation.

ASP.NET Core C# EF Core Repository Pattern

Full backend for job postings, applicant tracking, and user profiles — built with clean architecture (controllers → services → data access).

ASP.NET Core EF Core LINQ SQL Server

Real-time inventory tracking API with CRUD operations, database validation constraints, and optimized LINQ queries.

Python OpenAI Whisper XLM-RoBERTa HuggingFace

End-to-end NLP pipeline: YouTube audio extraction → Urdu transcription (Whisper) → English translation → sentiment classification (XLM-RoBERTa) → structured Excel output.

TensorFlow scikit-learn SVM KNN CNN

Comparative analysis of classical (SVM, KNN) vs. deep learning (CNN) on CIFAR-10, evaluated via accuracy, confusion matrices, and classification reports.

I'm open to internship opportunities, AI/ML collaborations, and backend projects. If you're building something interesting or want to discuss tech, feel free to reach out.