I don't just build things that work. I build things that can be trusted when they don't.
Autonomous state estimation, geospatial perception, grounded language architectures, and native desktop systems.
Designed and implemented a production-grade multiplicative Unscented Kalman Filter (UKF) for spacecraft attitude estimation using asynchronous multi-rate sensor fusion, adaptive fault detection, and Monte Carlo validation.
AI-powered geospatial platform fine-tuning Depth Anything V2 on satellite imagery to reconstruct metric Digital Surface Models (DSMs) from single-view optical imagery with interactive 3D WebGL flythrough and hydrologic flood simulation.
High-performance voice RAG pipeline combining Sarvam Indic STT, language-filtered FAISS + BM25 hybrid search, Reciprocal Rank Fusion, confidence gates, and grounded Groq LLM generation with latency benchmarking.
Built an embedding-based biometric iris recognition system using a custom CNN encoder and cosine similarity matching trained on the IITD Iris Database.
Distributed order lifecycle platform using Temporal durable workflows, deterministic event triage, FastAPI, PostgreSQL, and structured LLM reasoning to supervise long-running order state over days or weeks.
A production-grade, multi-agent AI system for deterministic and explainable e-commerce support decisions using RAG + rule engines.
An interactive, scroll-driven 3D scrollytelling frontend presenting the MTRA Mars terrain-risk research system using Three.js WebGL, Jezero Crater CTX DTM elevation terrain, and procedural Tone.js audio.
Binary sentiment classification system for Amazon-style product reviews supporting both DistilBERT and a bidirectional LSTM, deployed through an interactive Gradio interface.
A lightweight, always-on-top desktop companion built with Tauri 2 and React. Features transparent borderless windows, sprite-sheet animation, an idle/walk/react finite-state machine, and native OS tray integration.
Full-stack mobile auction platform built with Flutter, Node.js/Express, and PostgreSQL supporting authentication, real-time bid updates via polling, watchlists, and auction lifecycle management.
Web application that classifies satellite imagery into seven terrain classes using a CNN trained in TensorFlow and deployed via a quantized TensorFlow Lite model served through a Flask API.
I don't finish things halfway. If I take something on — a system, a problem, a stupid edge case at 2am — I see it through until it's right, not until it's good enough. I got into engineering because I wanted to understand how things actually work, not just that they do. That curiosity hasn't been selective. I've debugged telemetry on a nanosat, taught a model to read satellite imagery, and spent an unreasonable amount of time making sure a permission system couldn't be gamed by its own admin. The problems don't have to be related. They just have to be real. I build alone a lot. That means I've had to be the one who catches the race condition, writes the migration script, questions the schema at 3am, and cleans up the mess I made last week. I don't love every part of that — but I've stopped flinching at any of it.
I'm not sure where I'm headed exactly. But I know what kind of engineer I want to be: the kind who owns the work, asks the uncomfortable questions early, and never ships something they don't understand.
Still figuring out the rest.
NMIMS, Indore
Began B.Tech in Computer Engineering, building foundations in data structures, operating systems, and software engineering.
Servocontrols & Hydraulics India Pvt. Ltd.
Diagnosed and fixed a telemetry visualization bug in the EyeSat nanosatellite dashboard and refactored the telemetry pipeline into modular components.
Amazon
Selected among ~3,000 students nationwide for Amazon’s ML Summer School. Studied deep neural networks, generative AI & LLMs, causal inference, dimensionality reduction, and reinforcement learning through an intensive multi-week program.
Independent Work
Built a satellite image classification system using CNNs and deployed it. Also worked on NLP sentiment analysis models using DistilBERT and LSTM architectures.
Vyska Technologies Pvt. Ltd.
Developed backend systems for HRMS and LMS platforms including RBAC access control, optimized MongoDB queries, and reduced API latency.
Contract Project
Designed and deployed a MERN-based ERP system used by 500+ clients with automated inventory tracking, freight reconciliation, and role-based access control.
Google Cloud, Columbia Plus, Oracle
Completed multiple advanced programs including Columbia Machine Learning I and Google Cloud Generative AI Academy.