Good evening, I'm Justin
CS / DS @ UW-Madison | Incoming MSCS @ Georgia Tech
6 projects shipped • 3 internships • 3.76 GPA
Jump in
About me
Hey! I'm Justin Kim — a Computer Science and Data Science student at UW–Madison who builds and deploys full-stack systems and production backend infrastructure. I care a lot about understanding how systems actually work under the hood, from distributed services to model training pipelines.
I'm especially interested in machine learning, particularly LLMs and agentic systems. I've built language models from scratch to deeply understand their mechanics, and I'm a member of the WAISI AI Safety Fundamentals technical cohort, where I explore alignment and long-term safety in advanced AI systems.
I'm starting my M.S. in Computer Science (Machine Learning specialization) at Georgia Tech in Spring 2027, part-time alongside full-time work.
Last played on my Spotify
Fetched live from the Spotify Web API by this site's own backend — with Redis-cached tokens and automatic refresh-token rotation.
Spotify Wrapped
Year in Review
Four years of shipping — most recent on top
Spotify Wrapped
2026
Your #1 Role
SWE Intern @ Capital One | TRAIL Lab Researcher | Co-Founder @ MadLease
- 1Joined Capital One as a Software Engineer Intern — designed a GraphRAG pipeline that translates plain-English requirements into validated card fulfillment workflow configs, reaching 100% structural validity and 88% required-field completeness across 70 test configs and cutting build time from 2+ days to under an hour
- 2Built real-time graph validation (cycle detection, unreachable-state and invalid-reference checks) into a React/Monaco editor, adopted by 20+ intent engineers
- 3Joined UW–Madison's TRAIL research lab as an undergraduate research assistant, designing the tool-integration layer for an MCP-compatible multi-agent Q&A system with cross-tool retrieval and access controls, and red-teaming it against prompt injection and unauthorized tool use
- 4Scaled MadLease to 250+ accounts, 100+ listings, and 5+ building partners as co-founder and full-stack owner of the technical stack
- 5Moved image uploads to browser-direct S3 via presigned PUT URLs with retry/rollback handling across 1,000+ uploads, backed by CloudFront CDN delivery and Firebase UID/IP token-bucket rate limiting
- 6Boosted listing visibility across 25+ crawler types by serving structured metadata through Vercel Edge Middleware while routing real users straight to the React SPA
You started thinking long-term
Spotify Wrapped
2025
Your #1 Role
Full-Stack SWE Intern @ Entegral | RAG Engineer @ Denari | Co-Founder @ MadLease
- 1Led development of a production-ready crash photo capture and insurance intake platform at Entegral, replacing hard-coded crash-intake logic with configurable per-workflow documentation rules so providers could define claim requirements without code changes
- 2Shipped a company-wide reusable OCR module (Angular + Tesseract.js) that re-read VIN, license plate, and odometer fields in a background worker until each passed validation, reaching 90%+ accuracy
- 3Eliminated polling for estimate negotiation data by building a Pub/Sub database listener that propagated updates across multiple internal services in real time
- 4Built a citation-grounded tax RAG system for Denari indexing 50,000+ documents — answered 2,200+ CPA exam questions in ~30 minutes at 61% accuracy (outperforming GPT-4o-mini by 5 points and 10% on overall score), improving 14% in accuracy and 10% in citation quality over our first benchmark, all for under $5 in total compute cost
- 5Co-founded and deployed MadLease — a UW–Madison sublease marketplace — and fixed an IDOR vulnerability in the messaging service by adding conversation-membership authorization checks before allowing message reads
- 6Accepted Capital One's TIP Software Engineer offer for Summer 2026
You scaled production systems
Spotify Wrapped
2024
Your #1 Role
Machine Learning Engineer @ Astro AI Trading
- 1Promoted to Machine Learning Engineer at Astro AI Trading — led a 3-engineer team building sector-specific stock forecasting models that fused technical, macroeconomic, and sentiment features from 60+ daily news and social media sources, cutting held-out RMSE by 16% vs. a price-only baseline
- 2Fine-tuned a BERT 3-class news-sentiment classifier reaching 77% test accuracy, and built a separate BERT + PCA + KMeans clustering pipeline to route each article to its relevant market sector before scoring
- 3Built the first version of QuizClash at a hackathon — a real-time Kahoot-style platform that turns uploaded PDFs and YouTube videos into playable quiz games
- 4Scaled QuizClash to 40+ concurrent players with a WebSocket engine and a 5-stage ML pipeline (OCR, Whisper, T5, RoBERTa, Sense2Vec) for question generation
You started taking ownership
Spotify Wrapped
2023
Your #1 Role
Data Science Intern @ Astro AI Trading
- 1Started Computer Science and Data Science at UW–Madison
- 2Built and deployed FlixFinder — my first full-stack app — using Django, PostgreSQL, Scikit-Learn, and React
- 3Joined Astro AI Trading as a Data Science Intern and implemented preprocessing pipelines for live trading sentiment models
You started shipping real systems
Playlist
Top Projects
Justin Kim • 5 projects • 3 live to try • tap a track for the full story
Wrapped — Tech Edition
Your Top Genres
Click a genre to switch the breakdown below
Languages
Artist
Justin Kim
CS / DS @ UW-Madison
Liner Notes
This site is a project too — Next.js App Router, TypeScript, Tailwind, and Framer Motion, with a live Spotify Web API integration that caches access tokens in Redis and rotates its own refresh tokens so the player never goes stale.
Read the source