Computer Engineering @SJSU

Classifies incoming prompts, fetches real-time quotes from LLM providers, selects the lowest-cost option that still meets latency and quality policies, and settles usage with gasless USDC payments on Arc L1.

Built a graph-based deterministic equity strategy lab designed to shorten the path from trading idea to validated strategy, reduce overfitting risk, and keep experimentation cheap by testing multiple strategy variations side by side.

Built a mini app published on Alien.org where verified humans create crypto price predictions, stake WLD or USDC through MiniKit Pay, and build immutable reputation tied to World ID to make fake prediction spam significantly harder to game.

Developed a real-time object detection system using a Raspberry Pi and camera module, applying computer vision to identify and track objects efficiently for AI-powered edge applications.

Built a hackathon prototype using Reachy and edge AI to monitor study sessions in real time. The system detects distraction patterns such as phone usage or losing focus, tracks active study time, and reports session insights through a dashboard.

Built a robot-based parental control system at a Seeed Studio hackathon that enforced internet restrictions through a DNS-layer filtering setup. The project was designed to help parents manage screen time, block distracting or unsafe sites, and create a more interactive physical interface for digital safety. It combined robotics, networking, and system design into a working hackathon prototype.

Civic is a personal digital records hub that keeps life documents secure, categorized, and ready to share. It sends expiration reminders, offers one-tap Panic Mode for emergencies, and supports revocable, audited sharing links for trust-sensitive workflows.

Fine-tuned LLaMA 3 on a personalized dataset of personal notes and textbook excerpts to improve code-generation relevance, efficiency, and task-specific performance.

Python CLI that ingests PDF, TXT, and DOCX resumes and extracts key information. Combines regex-based contact detection, degree spotting, and skill matching against a configurable keyword list.

Turns Reddit and Twitter threads into shippable engineering work. Uses Gemini to detect recurring praise and complaints, cluster themes, and draft issue briefs with acceptance criteria so teams can act on real user feedback faster.

Served as lead developer on a Buildbox game and shipped it to both the Google Play Store and the Apple App Store.

Built a production-grade RAG platform for fast, accurate, context-aware search across enterprise documents by combining scalable ingestion, hybrid retrieval, fine-tuned LLaMA 3 responses, and secure APIs.

Enterprise-ready AI workspace that unifies team knowledge, conversations, and action tracking so organizations can execute faster with fewer tool silos.

Agentic content pack generator that turns messy notes, changelogs, and requests into a publish-ready update, announcement, and FAQ set.

Feedback-to-ticket bridge that turns high-signal X replies into GitHub issues, implementation plans, and shipped fixes.