Applied ML, agentic systems, retrieval, evals, and the occasional terminal rabbit hole.
I build and study applied ML systems, mostly around agentic AI, evaluation, retrieval, and production ML.
Currently: MSc Data Science at VIT, co-author of a recent arXiv paper on hybrid wavelet-based PINNs for portfolio management, and writer of Data Pe Charcha, where I try to explain ML without sanding off the interesting parts.
My default mode is simple: understand the system, build the smallest useful version, test it against reality, then make it less ugly.
- Agentic AI systems: planning loops, tool use, evaluation, and reliability
- Retrieval and RAG: hybrid search, reranking, dataset curation, failure analysis
- MLOps: FastAPI services, experiment structure, model evaluation, deployment basics
- Scientific ML: physics-informed neural networks, optimization, numerical methods
- Tools for myself: terminal apps, recommendation feeds, Neovim config, automation scripts
|
Hybrid wavelet-based physics-informed neural networks for derivative pricing under Merton jump-diffusion dynamics. |
Fraud detection project with standard MLOps structure and FastAPI deployment. |
|
A terminal Pomodoro timer in Rust with Vim keybindings. |
A personal content recommendation MCP server. |
|
Learning neural networks from the bottom up by working through micrograd. |
My Neovim setup. Still infant, but already opinionated. |
- Newsletter: Data Pe Charcha
- Website: mahaprasad003.github.io
- X: @mahaprasad_
- LinkedIn: mahaprasad003
I lift, read, write, and occasionally over-engineer my own productivity tools. I can still probably ruin a quiet evening with a deck of cards.