Industry and Research Work

Experience

Production engineering at Serafis (YC S25), full-stack internships, ML research, and program leadership with measurable impact.

Industry

2
Full-StackInternshipYC S25

Software Engineer Intern

Serafis (Y Combinator S25) · San Francisco, CA

Jun 2026 — Present

0-to-1 production React/TypeScript app as 1 of 3 core contributors

  • ▸Drove 0-to-1 development and launch of production React/TypeScript web app as 1 of 3 core contributors
  • ▸Built 2 Firebase auth flows, protected routes, and app shell with Vite, Tailwind/Shadcn, and TanStack Query
  • ▸Shipped 8+ product surfaces across Dashboard, Podcasts, Episodes, Companies, 4 AI Search tabs, and Ask AI
  • ▸Rebuilt search with hit-count relevance reducing discovery 12%, raising selections 8%, and cutting feedback 15%
  • ▸Integrated 10+ REST workflows for content-tag search, watchlists, 3-tier billing, PostHog, and AI citations
ReactTypeScriptViteTailwind CSSTanStack QueryFirebase
Full-StackInternship

Software Engineer Intern

Creative Labs @ UCLA ↗ · Los Angeles, CA

Mar 2026 — Jun 2026

TreeReq — UCLA course data pipeline and FastAPI/React integration

  • ▸Built Python scraping pipeline for UCLA course, prerequisite, and requirement data across 100+ majors/minors
  • ▸Mapped scraped catalog data into MongoDB schemas powering course dependencies and student-profile queries
  • ▸Designed FastAPI endpoints serving cleaned prerequisite data across 5+ React workflows for search and profiles
  • ▸Improved data quality with validation and deduplication logic across 1,000+ scraped UCLA course records
  • ▸Debugged React/FastAPI integration across 5+ release flows using GitHub PRs and typed API contracts
ReactFastAPIMongoDBPythonTypeScript

Research

1
ResearchML

Machine Learning Research Intern

ACM AI @ UCLA ↗ · Los Angeles, CA

Mar 2026 — Jun 2026

Chess transformer fine-tuning and interpretability research

  • ▸Fine-tuned GPT-style chess transformers on 10,000+ Lichess positions with PyTorch/Hugging Face
  • ▸Reduced transformer training time by 30% by integrating DeepSpeed distributed training into PyTorch pipelines
  • ▸Built PyTorch evaluation pipelines benchmarking move prediction across 3+ models and 1,000+ test positions
  • ▸Decoded board features across 10+ transformer layers by developing custom probing pipelines for interpretability
  • ▸Analyzed 1,000+ model predictions with Hugging Face probes to study chess decision patterns and failure modes
PythonPyTorchHugging FaceDeepSpeed

Programs

1
ProgramLeadership

Project Manager Intern

NASA L'Space ↗ · Remote

Sep 2025 — Dec 2025

Led 10+ contributors through NASA Phase I mission design

  • ▸Delivered 2 NASA-reviewed Phase I reports by leading 10+ contributors across scope, milestones, and delivery
  • ▸Led 3-team integration by translating MATLAB trajectory, propulsion, and payload analyses into requirements
  • ▸Improved doc throughput 25% and cut compile time 30% by standardizing Git, LaTeX, and Python workflows
MATLABPythonLaTeXGit