My Blog
Stories about things I have learned, projects I am hacking on,
and discoveries worth sharing.
How I Built a 14B Parameter Avatar System That Runs at 20 FPS
A technical deep dive into deploying Alibaba's WanS2V-14B model on 5x H800 GPUs for real-time avatar generation, covering the streaming pipeline, LoRA fine-tuning, and inference optimization.
Voice AI for 31 Million Citizens: Building Sarathi
How we built the Sarathi Voice Agent to deliver government services in Assamese and Bodo using faster-whisper ASR, Bhashini TTS, and sub-second latency optimization for kiosk deployment.
Building Production-Ready RL Agents
A deep dive into creating reinforcement learning agents that work in production environments, with lessons learned from the Fundora project.
RAG Systems That Don't Hallucinate: Engineering Zero-Trust AI
How we engineered the Aya Knowledge Base with grounding scores, provenance tracking, and hallucination detection to build enterprise RAG that clients actually trust.
Shipping AI Agents for Enterprise Clients
How we built and deployed vision-aware AI agents and real estate automation systems at AlysAI for enterprise clients.
The MCP Protocol Will Change How AI Agents Work Together
What the Model Context Protocol is, why it matters for multi-agent systems, and what I learned building MCP-compatible tools at the Lost & Found hackathon.
MLOps on Kubernetes: A Practical Guide
Setting up a complete MLOps pipeline on Kubernetes with auto-scaling, model versioning, and monitoring.
Why I Left Web3 to Build AI: A Founder's Journey
A personal reflection on transitioning from building Web3 products at MagicSquare and LunarCrush with 190k+ users to AI engineering and founding Sarathi Studio.