I'm Aaditya (Icey), a Software & MLOps Engineer based in India, specializing in low-latency distributed systems, real-time MLOps pipelines, and high-performance native applications.

I build fault-tolerant systems and production-grade AI infrastructure, and I actively compete in technical hackathons.

• 🛡️ Sentinel: Real-Time MLOps Anomaly Detection Engine: An enterprise-grade, sub-15ms streaming financial fraud & outlier detection engine.
  • High-Throughput Streaming: Ingests high-frequency transaction streams via Redis Streams with consumer groups and Dead-Letter Queue (DLQ) fault tolerance.
  • Low-Latency Inference: Quantized a PyTorch Autoencoder to ONNX Runtime Dynamic INT8, driving single-sample CPU inference down to <0.1ms.
  • Benchmarked SLA: Measured p95 end-to-end pipeline latency at 2.09ms (beating the strict <15ms SLA) at 540+ RPS load with 0% failure rate.
  • Automated Drift Engine: Monitors statistical data drift using Evidently AI (2-sample KS-tests & Wasserstein distance) with zero-downtime hot-reloading.
  • Observability & UI: Containerized multi-stage Docker stack exposing Prometheus metrics, Grafana dashboards, and a live React 19 telemetry UI.

• 🏎️ Silent Co-Driver (4th Place - GrandPrix AI Finals): An enterprise-grade, edge-to-cloud F1 pit-wall dashboard. Architected a hybrid pipeline using Node.js, Python (Librosa DSP), and Gemini 3.7 Flash Multimodal AI to extract real-time vocal stress biomarkers and generate tactical directives.
• 📍 Lumina / CampusFind (1st Place - GDG Hackathon): An AI-powered lost-and-found portal. Integrated Gemini AI and TomTom Maps with a React/TypeScript/Firebase stack for real-time geolocation and multimodal missing-item matching.

• 👾 KOrunner-Android: A high-performance infinite runner engineered natively in Kotlin. Bypassed heavy third-party engines to build a custom game loop and AABB collision physics engine directly on Android's SurfaceView for a locked 60 FPS.
• 🎮 Pokemon-Trainer-Legacy: A retro-style Pokémon battle engine built from scratch in Python/Pygame, featuring complex Object-Oriented Programming (OOP) for type-effectiveness math and enemy AI move calculation.
• 🛡️ Codev-Viewkey-Launcher: A privacy-first, lightweight URL launcher for Android built with Jetpack Compose.

• 👁️ Focus-Monitor: A real-time attention tracking pipeline built with OpenCV and MediaPipe. Computes facial landmarks and head pose estimation to calculate user focus levels for proctoring environments.
• ☁️ Full-Stack Weather App: A complete system bridging a native Kotlin Android frontend with a custom Python/Flask REST API backend.

Languages     : Python 3.11+, Kotlin, TypeScript, C++, SQL, Bash
MLOps & AI    : PyTorch, ONNX Runtime, Evidently AI, MLflow, OpenCV, MediaPipe, Scikit-Learn
Backend & Ops : FastAPI, Asyncio, Redis Streams, Docker, Docker Compose, Prometheus, Grafana
Frontend & UI : React 19, Jetpack Compose, Tailwind CSS, Recharts, Firebase

🚀 Always building, optimizing, and looking for the next technical challenge. Open to new opportunities and collaborations!

• Portfolio: icey-portfolio.vercel.app
• LinkedIn: in/icey07