Divith Selvam  Pronounce my name

I build the systems that keep large language models running in production: the part that decides which machine answers a request, and the part that lets an AI agent finish a long job instead of starting over every time something breaks.

• Software engineer and AI engineer working on LLM systems, agents, and inference. AI and Data Science graduate, 2026.
• A distributed LLM serving control plane in Go, Python, vLLM, and Kubernetes. It routes streaming requests by tenant quota, replica load, and KV-cache state, with backpressure, circuit breakers, and OpenTelemetry tracing.
• A durable agent runtime for MCP and computer-use workflows, with event-sourced state, checkpoints, retries, and crash recovery, plus scoped credentials, permission checks, and human approvals before sensitive actions.
• Tool-use post-training and LLM evaluation with PyTorch, Transformers, PEFT, QLoRA, and TRL, including deterministic tool-use environments and synthetic trajectories.
• At CHN Technologies I work on Django and NestJS services, PostgreSQL schemas and indexing, and Docker CI/CD. Before that I built Blazor and .NET components and profiled SQL performance at Wise Work.
• I founded OneDot Communications. Open to full-time software engineer and AI engineer roles, remote or relocating.