About Me
I build full-stack AI products end to end. FastAPI backends, vector retrieval pipelines, React Native apps, and Next.js interfaces. I founded Bridge, run Applied Engineering @ SJSU, and research decision-making because understanding how people fail is the only serious way to design systems that prevent it.
Data Science @ SJSUWhere I’ve worked
I build full-stack AI products end to end. FastAPI backends, vector retrieval pipelines, React Native apps, and Next.js interfaces. I founded Bridge, run Applied Engineering @ SJSU, and research decision-making because understanding how people fail is the only serious way to design systems that prevent it.
Data Science @ SJSUKey decisions
Bridge had events, free food, social discovery. Tested well. Cut all of it. The real problem was wrong information at the wrong time, not information overload.
Talked to students, collaborated with SJSU's ISSS office, ran structured surveys. The product changed completely from the first version.
The obvious approach was to build from what a user owns. We rejected it, it created friction, didn't feel impressive. Started from taste instead.
Applied Engineering finds problems companies have sidelined because of deadlines. Those are the real problems. That's how we ended up with Bloom Energy.
21,800 Reddit posts. Three domains. Quantitative modelling. Not research : a design tool for Bridge's risk intelligence layer.
Projects
Featured builds, case studies, and experiments.
Three-signal purchase intelligence system built for Phia. Solves cold-start personalisation, session intent inference, and return-rate prediction in fashion. Taste extraction via FashionCLIP (ViT-B/32, fine-tuned on 800K Farfetch items) from Pinterest boards at zero purchase history. Session intent inference from browsing coherence. Wardrobe-integrated purchase confidence scoring. Results: +102% taste relevance vs. random baseline · +52% vs. popularity ranking at zero saves · 69.5% purchase confidence accuracy vs. 50% random. Agentic stylist: multi-turn Claude Sonnet 4 with 4 live tools, SSE streaming mixed content, FAISS IndexFlatIP over 2,364-item catalog, MMR reranking (λ=0.7).
Founded to help students navigate high-stakes decision windows before they close. Identified the problem from personal experience. Ran structured research with 200+ students and SJSU's ISSS office. Built a stage-aware engine that surfaces the next critical action per user journey. Killed the social features when they diluted the core. The insight: one missed step — a CPT deadline, an OPT window — can change your entire trajectory. Bridge's risk intelligence layer flags these before they close, not after. Shipped 25+ features across web and mobile. Early access waitlist live.
Led 3-engineer team to design and ship a production RAG customer service agent for SGConsulting. RAGAS evals: faithfulness 0.83 · answer relevancy 0.81 · context recall 0.89. LangSmith tracing: 5.25s average latency · $0.0065/query. Dockerized with GitHub Actions CI/CD, pytest suite, and zero-downtime releases. Technical runbooks adopted by client engineering team.
Analysed 21,800 Reddit posts across career, relationships, and education to model when and why people reverse high-stakes decisions. Key findings: regret peaks differ by domain — career regret surfaces fastest, relationship regret surfaces latest. Inaction regret outweighs action regret 2:1 in long-horizon decisions. These findings directly inform Bridge's risk intelligence design, the system is built around the actual timing of when decision windows close in real student journeys, not assumed timelines.
Work
A timeline of roles, internships, and collaborations.
Sep 2025 – Present · San Jose, CA
Co-founded and run SJSU's first student-run software agency: 100+ members, live production work for industry clients. Act as PM and tech lead across engagements: interview companies to find underserved problems, scope projects, recruit and manage cross-functional teams, own delivery end-to-end. Current clients: SGConsulting, Bloom Energy.
Aug 2025 – Present
Identified the problem from personal experience. 200+ interviews. Killed the social features when they diluted the core. Built stage-aware engine + mobile app. Currently in early access.
May 2025 – Aug 2025 · San Jose, CA
Built Python data pipelines covering 20+ RF device categories with 100% dataset accuracy. Refactored legacy validation modules cutting test setup time 33% and throughput +25%. Designed structured validation schemas reducing review cycles from 3 to 1; changes adopted across the full RF test suite in production. Cut engineer onboarding time ~40%.
Jul 2024 – Sep 2024 · Remote
Built AI-powered applications across a structured 7-week fellowship. Shipped a smart pantry management app using Next.js, Firebase, and TensorFlow.js with image recognition and recipe generation, improving inventory tracking efficiency 50%. Built RecMyProf: an AI professor recommendation system using Next.js, OpenAI API, Pinecone, and FastAPI, rating and matching 300+ professors with semantic search.
May 2024 – Aug 2024 · Chennai, India
Launched full B2C e-commerce website from design to deployment — 50+ hours of design, development, testing, and deployment using HTML, CSS, JavaScript; improved SEO by 90%. Built an inventory management system in React + Firebase and a billing application in Django that increased operational efficiency 60% and reduced database query latency 30% via indexing.
