Distributed ML Training & Inference Platform
End-to-end ML training pipeline with LoRA fine-tuning, 8-bit quantization, and Kubernetes orchestration for GPU scaling. Designed for cost-efficient model iteration.
Open to Senior SWE / AI Engineer roles
Senior Software Engineer · AI Platforms
I build scalable distributed systems, microservices, and production ML platforms across AWS, Azure, and GCP — from model-serving pipelines on Kubernetes to agentic RAG systems and full-stack GenAI apps.
01 — Experience
Education
BTech, Electronics & Telecommunications
Vishwakarma Institute of Information Technology (SPPU, Pune)
2020 - 2024
CGPA 8.69
02 — Stack
Languages
Backend
Frontend
Cloud & DevOps
Data & Messaging
AI / ML
Observability
Tools
03 — Projects
Systems I've architected — ML infrastructure, RAG pipelines, and event-driven services.
End-to-end ML training pipeline with LoRA fine-tuning, 8-bit quantization, and Kubernetes orchestration for GPU scaling. Designed for cost-efficient model iteration.
Event-driven full-stack system with real-time WebSocket streaming, agentic workflows, and Docker Compose plus Nginx architecture built for high availability.
AKS and Kubernetes serving stack processing 50K+ inference requests daily with sub-200ms latency. Instrumented with Grafana and Prometheus observability.
Production RAG system for enterprise document retrieval, improving retrieval accuracy by 35% through systematic benchmarking of chunking and embedding strategies.
Real-time data ingestion pipeline across distributed systems using Kafka and RabbitMQ. Handles high-throughput streams with backpressure and DLQs.
Optimized data pipelines using Numba and CUDA, achieving 4x throughput improvement over the CPU baseline for numerical and inference workloads.
04 — Writing
Lessons from architecting production ML serving on AKS and hitting sub-200ms latency without burning through GPU budget.
Six months of production RAG taught me that retrieval quality dominates, not the LLM. Here is how we improved accuracy by 35%.
Both are great. Neither is universal. Here is how I decide when designing event-driven microservices.
05 — Contact
Hiring, collaborating, or just want to talk ML infra? My inbox is open.
Mumbai, India