Software Engineer · Machine Learning
I build software that stands the test of time. My focus is machine learning grounded in mathematical rigor, from equivariant neural architectures to the infrastructure that deploys them.
ML & AI
Languages & Libraries
Engineering & Infra
Research & full-stack projects
Probed a frozen ViT-B/16 layer by layer to reveal a spatial hierarchy: patch boundaries become linearly decodable at L5 (AP = 0.833), depth at L8 (MAE = 0.0875). Causal ablation shows depth is concentrated in a single direction out of 768 — erasing it degrades the signal by 165%, while erasing any other direction changes it by less than 1%.
A multi-agent home energy system where LLM-powered house agents negotiate with a neighborhood coordinator to reschedule shiftable loads. Over 10 simulated days, peak grid demand fell by 45%, electricity costs by 42%, and grid CO₂ by 30% — with no hardware changes and private household preferences never exposed to the coordinator.
A multi-agent RAG system for compliance that reads live SEC filings and internal documents, then answers questions through a LangGraph-orchestrated pipeline of six specialized agents. Every report is extracted, scored for impact, and self-evaluated for faithfulness before delivery — grounded in real sources, not training memory.
Full-stack investment dashboard for managing stock watchlists with interactive price charts, document aggregation from SEC EDGAR and Polygon.io, and Google OAuth. Multi-layer caching with React Query.
Full-stack recipe discovery app that matches meals to ingredients you already own. Ranks suggestions by ingredient match score, dietary restrictions, and estimated cost using live Kroger grocery pricing.
Java-based image processing package with a Swing GUI, built on strict MVC architecture and the Command design pattern. Supports RGB channel splitting, sepia filtering, histogram visualization, compression, downsampling, and masked partial manipulation — all via a pluggable command pipeline.
Professional & academic roles
Jan 2025 – Present
Boston, MA
Northeastern University
GTA for Object-Oriented Design, supporting 400+ students through weekly code reviews, lab sessions, and office hours. Teaching SOLID principles, design patterns, and clean architecture, translating academic concepts into production-ready habits.
Jun 2022 – Jul 2024
Mumbai, India
Oracle
Built an NLP-based bug assignment system that cut manual triage time by 35%, routing defects to the right teams automatically. Led microservices automation on OpenShift improving deployment efficiency by 40%.
Automated patching across 180+ OCI servers reducing manual errors by 30%. Designed CI/CD pipelines with Jenkins and Groovy, standardizing release workflows across multiple microservices teams.
Education
MS Computer Science · GPA 4.0/4.0
2024 – 2026 · Boston, MA
Geometric Learning, Actionable Interpretability, ML, NLP, Scalable Distributed Systems, Algorithms
B.Tech Information Technology · GPA 8.21/10
2018 – 2022 · Mumbai, India
Long-form technical essays
Nov 2025
Geometric DL
Most neural networks learn symmetries through brute-force augmentation. Equivariant architectures encode them directly using group theory, producing networks that are smaller, faster, and provably consistent under transformations.
Feb 2026
Sequence Modeling
The headline numbers say Transformers win. But when you examine compute budgets, long-range dependencies, and inference efficiency, the picture becomes far more nuanced.
Mar 2026
Operations Research
Classical operations research assumes static environments. Combining Q-learning with metaheuristic optimization for stochastic demand yields a 90% cost reduction, but the path there is full of surprises.
Apr 2026
Mech. Interpretability
Vision Transformers trained only on class labels never see depth or boundaries. Yet probing them layer by layer reveals an actively maintained spatial hierarchy — and a single direction in 768 carries the depth signal.
I'm looking for software engineering and ML roles where I can build systems that last. Currently based in Boston.