Final-year Computer Engineering student at SPPU, based in Pune. I like taking ambitious technical ideas from a rough mental model to a working, observable product.
My work sits at the intersection of software systems, machine intelligence, and thoughtful product design — from GNNs that find coordinated fraud to small models fine-tuned on consumer hardware, and agents that operate inside real cloud environments.
I’m especially interested in the space between a model that works in a notebook and a system that can be trusted in the world.
WEB & SYSTEMS AI / ML CLOUD INFRASTRUCTURE PRODUCT ENGINEERING
|
Graph-neural-network fraud detection for UPI payment graphs. Finds mule accounts and coordinated fraud rings that transaction-by-transaction rules miss. Includes ring-aware evaluation, GNN explanations, drift monitoring, and a FastAPI dashboard.
|
An AI SRE agent for real cloud environments. Collects infrastructure telemetry, investigates root causes, surfaces cost waste, forecasts operational problems, and alerts only when an action matters.
|
|
An Android volume controller driven by movement. Uses GPS as the authoritative vehicle-speed signal, with a self-calibrating step-sensor fallback for walking and running. Built with platform APIs and hardened for foreground execution, lock-screen use, and OEM battery policies.
|
A compact Gemma 2B instruction model. Fine-tuned with Unsloth on consumer hardware, focused on making useful model experimentation accessible without a large training budget.
|
|
Flood-risk forecasting for Maharashtra's river basins. A GRU sequence model that predicts flood risk 24–72h ahead over the Godavari, Krishna & Konkan basins — upstream-lag spatial features, isotopic calibration (4 risk tiers), and baseline ablation against a stacking classifier.
|
| Build React · Node.js TypeScript · Laravel |
Learn PyTorch · PyG Hugging Face · Unsloth |
Ship Cloudflare · AWS Oracle · FastAPI |
Think about Systems · Agents ML products · Reliability |
understand the problem → build the smallest useful system
↑ ↓
study the failure ← measure what actually happened
I care about the parts that are easy to skip: clear assumptions, meaningful evaluation, failure modes, observable behavior, and documentation that tells the truth about what a project can and cannot do.
