aswinkumar

Aswin Kumar — Full-Stack & AI/ML Engineer
I build production systems that hold up under real load — not demos. Currently a Full-Stack Developer Intern at SMIE Industries, architecting real-time logistics and tracking systems in production.
My strongest work centers on distributed, event-driven infrastructure: NightWatch Atlas, a real-time intelligence platform running 6 containerized microservices on Kafka, sustaining 10–50 events/sec at sub-2s latency with a hybrid PostgreSQL + Elasticsearch storage layer — and F1 Telemetry Anomaly Engine, combining Kafka streaming, async FastAPI inference, and a PyTorch autoencoder with explainable anomaly detection, replayed live at 50Hz.
On the AI side, I've built a 5-stage RAG pipeline (Llama 3.1-8b, FAISS, HuggingFace Transformers) for evidence-based causal reasoning over unstructured text.
I've also contributed to open source at scale — a merged PR to PyTorch (Meta AI) fixing Python 3.12 type-checking and CI/CD failures — proof I can navigate large, unfamiliar production codebases without hand-holding.
Core stack: Python, TypeScript, C++ · FastAPI, Kafka, WebSockets · PyTorch, HuggingFace, FAISS · PostgreSQL, Redis, Docker, AWS.