PinnedInCodeToDeploybyArijit Sengupta·Apr 29Building a Durable MCP Agent in Conductor -conductor-ossNetflix Conductor is now Conductor OSS.
PinnedInCodeToDeploybyArijit Sengupta·Apr 10Create a Custom MCP Server for Codex with TypeScriptA practical step-by-step guide for developers.
Arijit Sengupta·1d agoDoes KV Caching Really Speed Up LLM Inference?Understanding prefill, decoding, GPU memory, and inference latency by benchmarking Qwen2.5–0.5B with KV caching enabled and disabled.
InTowards AIbyArijit Sengupta·Nov 1How to Fine-Tune Qwen2.5 with LoRA and Publish Your Model on Hugging Face: A Step-by-Step GuideBuilding a fashion attribute extractor with teacher-generated data, a 0.5B student model, and Google Colab.
Arijit Sengupta·Sep 27Can Agent Simulation Catch Regressions That Smoke Tests Miss?I introduced a small bug into an AI travel agent. Five ordinary smoke tests passed. Multi-turn agent simulations caught the regression in…
InStackademicbyArijit Sengupta·Aug 30Node.js API Can Be Async and Still Block: The Hidden Cost of JSON.parse() and JSON.stringify()Node.js is often described as fast because of its non-blocking, asynchronous architecture.
InGoogle Cloud - CommunitybyArijit Sengupta·Aug 16How I Built Stale Lease Recovery for Async Gemini Jobs on Cloud RunPROCESSING does not always mean a worker is alive
InStackademicbyArijit Sengupta·Aug 15How to Build an Async Gemini Job Pipeline with Cloud Tasks and Cloud RunBuilding a background GenAI workflow with Cloud Run, Cloud Tasks, Firestore, Vertex AI, and an idempotent worker.
InStackademicbyArijit Sengupta·Aug 8Why Inference Optimization Isn’t One-Size-Fits-AllTraining an AI model is only half the challenge. The other half is serving millions of requests efficiently.
InCodeToDeploybyArijit Sengupta·Jul 12The Complete Guide to Securing Model Context Protocol (MCP)Authentication, Authorization, Transport Security, and Production Best Practices