Latest Video

2026

Trained NVIDIA TFM from scratch on 2×RTX 4090 — Trained a 29M-parameter Llama decoder model on 24M financial transactions in 18 minutes. Extracted embeddings and achieved +32% AP lift on fraud detection when combined with XGBoost. Blog

Exploring idea on AI in Marine Industry - Exploring the idea of using AI to optimize shipping routes and reduce fuel consumption in the marine industry.

Tailscale Browser Extension - Contributed Windows support and dependency updates to tailscale/ts-browser-ext — Blog
Talk on .claude/ folder - Gave a internal talk in Thoughtworks on how to use the .claude/ folder - Blog

Karpathy’s AutoResearch — Tried to implement in nvidia/cudf library - Will write Blog post on it soon!
HomeLab Update - Moved from Homepage to Glance because Read More
Launched personal site — Built a new portfolio site How it is designed
SelfMemory Talk - Gave a virtual talk on SelfMemory at Paavai College of Engineering - Certificate Slides
Explore SmythOS - Build agents for Windows Laptop - Will write Blog post on it soon!

Started itwillSync - Mirror any terminal-based coding agent to your phone over LAN — open source, zero cloud, agent-agnostic. Blog
Langchain Community Talk - Gave a talk on “Building AI Agents: From Concept to Production” at the Langchain Community event, covering agent design, development, and deployment best practices. Photos

Built a HomeLab - Build a HomeLab with Intel Pentium Machine - Maybe I should write a blog post on it?

Build Travel MCP - Build a MCP Server which is plan the travel plans with multiple modes of transport with flights, trains and buses. - Demo

SelfMemory — Open Source AI Memory Protocol (100+ active users). Designed cross-model Memory Management Protocol (MMP) enabling persistent conversational context across LLM providers. Built SDK + MCP integration layer supporting agent interoperability. selfmemory.com

TWCline — Built a public-facing coding agent (VSCode extension) for both internal and external users, with centralized metrics tracking and unified coding standards

SelfHosting DeepSeek — Successfully self-hosted DeepSeek V3 on AWS bare metal. This is intersting story because I didnt have access to H200 Machine I tried doing with A100 machine. Story here

LLM GPU Calculator — Tool for calculating GPU requirements for LLMs (1000+ visitors/month)

AI Engineer at Thoughtworks — Joined full-time after internship
“Qira AI” — Designed and implemented AI backend for a Windows-based application controlling laptops for the world’s largest laptop manufacturer
Orchestration for E-Commerce — Developed orchestration for a customer-facing laptop management application
LLM Integration for GPU Manufacturer — Integrated client’s own LLM into existing biomedical applications, demonstrated value-add through modeling and capability testing
MCP Server for E-Commerce — Built an HTTP Streamable MCP server for placing & searching orders for India’s largest e-commerce store
AIFSD Enablement — Enabling AI-first software delivery organization-wide, optimizing the whole SDLC from Product Owner to Quality Analyst for a global biopharmaceutical company
LLM Hosting — Custom model deployment on AWS Bedrock and self-hosted DeepSeek V3 on bare metal

Synthetic Dataset Generation — Built tooling for generating synthetic datasets

AI Research Intern at Thoughtworks
Embedding Model Training — Trained Sentence Transformer model for RAG system to improve document retrieval
Computer Vision Transfer Learning — Implemented transfer learning on YOLO for object detection, deployed on Jetson AGX Orin Nano for real-time industrial safety monitoring
Multiagent Systems for SDLC — Automated the SDLC using Autogen, from requirement gathering to documentation
POCs — Visual QA, LogBot, Text2SQL, Agentic Customer Care agent

PPE Detection DL Models — Trained object detection model for industrial safety using TensorFlow and YOLOv8. Developed and deployed a real-time model to identify workers wearing safety equipment

B.Tech in Artificial Intelligence and Data Science — BIT - Bannari Amman Institute of Technology