Undergraduate at the University of California, San Diego, double majoring in Mathematics–Computer Science & Data Science. I research AI memory models and intelligent virtual assistants, aiming to build systems that remember and adapt over months, not minutes. Beyond research, I co-founded Falcon Eye — a synthetic-vision platform for aircraft that raised $1.2M and is being evaluated for integration into the HAL Tejas, India’s flagship 4.5-generation fighter jet. I bring the mindset of a builder-researcher-entrepreneur with a focus on real-world impact. Outside of work, you’ll usually find me reading or sketching ideas at Geisel Library, 2nd Floor.
A terrain-aware synthetic vision system built in C++/OpenGL, delivering real-time 30 FPS rendering on <$100 hardware. Designed for robustness and frame-rate stability with stitched elevation data and GPU acceleration.
Recently secured $1.2M in funding and selected for potential integration into the HAL Tejas 4.5-generation fighter jet. See more details in the Experience section.
Ongoing research with Prof. Julian McAuley and Yu Wang on how large language models can maintain long-term conversational memory. We are experimenting with memory update algorithms that balance retention, eliminate redundancy, and resolve conflicts. The prototype integrates Redis, hybrid retrieval, and GPT-based enrichment, and is being prepared for an ACL 2026 workshop submission.
See more details in the Research section.
Strata is my YC-ready productivity assistant that brings Gmail, Calendar, and Tasks together into one unified experience. It’s built to handle everyday workflows with offline support and conflict-free sync across devices.
The project is currently under active development and is being shaped into a startup application. See more in the Projects section.
ACL 2026 Workshop (in preparation)
A memory layer for multi‑session dialogue that blends dense retrieval with heuristics and GPT‑based enrichment. Implements decay, merge, and conflict‑resolution policies for stable long‑term context.
Advisors: Julian McAuley, Yu Wang.
Designing and evaluating memory update algorithms for LLM-based chat systems, focusing on balancing retention, redundancy elimination, and conflict resolution. Implemented the LLM-Memory-Engine prototype integrating Redis, GPT enrichment, and retrieval strategies to simulate long-term memory. Manuscript in preparation for ACL 2026 Workshop.
Advisors: Emma Waldron Chen, Ross Ingram.
Developed a production-ready long-term memory system using vector databases and heuristic scoring. Deployed internally across 500+ daily users, yielding a 30% improvement in measured helpfulness. Research remained private and used within the team.
Independent Research Project.
Modeled spatiotemporal trajectories of baseball home runs using Statcast data. Built ML pipelines to predict landing zones and generate stadium heatmaps. Validated against live MLB data; targeting submission to SABR Analytics Conference.
Advisors: Suresh Shakkarwar, Mandar Kulkarni.
Investigated outlier detection strategies for large-scale property datasets (10,000+ U.S. records). Implemented Isolation Forest and statistical thresholding, improving model accuracy by 18%. Work remained internal and was adopted within Diamond Realty Investment (Mitsubishi Corporation) workflows.
Experiments with decay, merge policies, and conflict resolution under noisy user histories.
Early metrics and ablations for retrieval quality and downstream task accuracy.
C++ / OpenGL flight-assistance system
Built terrain-aware synthetic vision rendering at 30 FPS on <$100 hardware; secured $1.2M in funding; interest from HAL Tejas fighter jet program; emphasis on robustness and real-time guarantees.
Cross-platform productivity assistant
Strata is my take on a YC-ready assistant. It keeps email, calendar, and tasks in sync across devices so you can plan your day, handle an inbox, or reschedule meetings even offline. It’s designed for daily use, with fast capture, unified search, and conflict-free sync once you reconnect.
Strata brings Gmail, Calendar, and Tasks together in one assistant that actually works across platforms. You can jot down tasks on your phone, update calendars on your laptop, and everything stays consistent—even offline. It’s built for people who live in multiple apps but want a single assistant they can trust every day.
ShardWise is a system for running jobs reliably across a cluster of machines. It’s built for research pipelines, ETL jobs, or large-scale computations that must continue even if one node fails. By handling leader election, replication, and sharding, ShardWise ensures work never stops and results are consistent.
Orion is a desktop AI assistant for macOS that helps automate everyday workflows. It can open apps, file documents, schedule events, or draft quick messages. The goal is to bring the convenience of an AI assistant directly into the Mac ecosystem while keeping data private and on-device.
Sentinel tracks web pages and only alerts you when the actual meaning changes. If a pricing page, policy document, or research dataset updates, you’ll know about it—but it won’t bother you when buttons move or CSS changes. It makes monitoring information sources practical without drowning in noise.
javelinLang is a small language I built to make repetitive desktop tasks scriptable. Instead of manually clicking through a UI or renaming hundreds of files, you can write short scripts that do it for you. It’s especially handy for QA testing, report generation, or automating apps that don’t have command-line tools.
Applied AI overview for misinformation detection with practical examples.
Concise tutorial covering core ML concepts and algorithmic foundations.
A personal intro: what I study, what I’m building, and how I think — from late nights at Geisel to research on AI memory and assistants, plus the projects that keep me honest.
Short technical updates on decay/merge policies, redundancy pruning, and evaluation in multi‑session dialogue.