Pune, India — available for interesting problems
I'm Mahesh Paul, a Data Scientist & AI Engineer specializing in LLM-based multi-agent systems, deep learning, and end-to-end PyTorch pipelines.
A builder who likes the space between research and production.
Jan 2024 — Jun 2024 · Pune, Hybrid
MIT World Peace University, Pune
MIT World Peace University, Pune
Advanced past college-level selection rounds.
Python Pro Bootcamp · Machine Learning A–Z · Full Stack Web Developer
Experiments, products, and systems built to move an idea from notebook to useful.
Four collaborative LLM agents — News, Trading, Risk, and Manager — produce BUY / SELL / HOLD signals with retrieved context. Backtested on GOOGL at 127% return and 2.25 Sharpe.
A 3-layer Graph Attention Network trained on 46,564 Elliptic Bitcoin transactions. Inverse weighting handles imbalance, while attention weights make the predictions explainable.
A few-shot pipeline for undeciphered Indus Valley symbols, meta-trained on Omniglot and tested with only 1–5 examples per class. Includes a live Streamlit interface.
Multi-class detection across 2.5M+ records and 11 attack categories. A voting classifier ensemble reaches high accuracy while keeping inference fast enough for live traffic.
A practical stack for learning fast, shipping clearly, and measuring what works.
Have a difficult dataset, a fuzzy product idea, or a system that needs a sharper brain? Send a note.