Software engineer based in Bangalore, India.

Most of what I work on is building Agentic AI systems.

• LLM Systems (SWE Agent): Coding agent built from scratch without frameworks. Features a ReAct loop, context compaction, and custom tool runtime. Scored 26 out of 30 on SWE bench Verified.
• MedChain: RL environment for hospital supply logistics built for OpenEnv India 2026.
• Variational Autoencoder on QuickDraw: Trained a VAE from scratch on Google QuickDraw to generate sketches.
• Bhagavad Gita Conversational AI: Fine tuned Gemma 3 1B on a dataset assembled via book OCR, web scraping, and YouTube transcripts.
• Shopmate: LangGraph shopping assistant that clarifies requirements first, then compares items from Amazon and Flipkart.
• Election Portal: Campus election platform for IIT Roorkee handling 10K plus student votes in a 2 hour window on AWS.
• Redis From Scratch: In memory key value store built in C with TCP socket handling, RESP protocol parsing, key expiry.

• Guide to train on multiple GPUs using DDP
• Variational Autoencoder (VAE): How to train and inference?
• Overview of Wan 2.1: Text to Video Model
• How does Node.js work behind the scene?
• Bits on the Move

nikm5502@gmail.com | Portfolio | CV | LinkedIn | X | Medium