Software engineer based in Bangalore, India.
Most of what I work on these days is systems around agentic AI workflows, local LLM serving, and the backend infrastructure that goes with it.
Also spending time on world model research, reading papers (on surface). Checkout github.com/nik-55/world-models.
- Redis From Scratch: Built a lightweight in-memory key-value store in C. Covers TCP server, RESP protocol parsing, TTL support, and master-replica synchronization.
- Shopmate: Multi-step LangGraph agent that automates product discovery across Amazon and Flipkart, with intent understanding and recommendation scoring.
- Variational Autoencoder on QuickDraw: Trained a VAE from scratch on the Google QuickDraw dataset. ResNet-style architecture, KL annealing, custom inference using aggregate posterior sampling.
- Bhagavad Gita Conversational AI: Fine-tuned Gemma-3-1B on a custom dataset built from web scraping, OCR on scanned books, and YouTube audio transcription.
- Election Portal: Official campus election platform for IIT Roorkee, used by over 10K voters. Built with Next.js, Django, PostgreSQL, and AWS.
- IKS Console: Certificate generation platform built for the Indian Knowledge Systems Division, Ministry of Education, Government of India.
- 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 | bio.nik-55.dedyn.io | LinkedIn | X | Medium