andrew ( jungmin ) kim

student / software dev / designer / vfx / web3

korean-american computer science senior at virginia tech with a passion for design, web3, ai/ml, and creative coding. i love competing in hackathons and working on random projects for fun. i also play tetris competitively (ranked #197 in the world)

projects

3dera

3dera transforms documented history into immersive vr environments by orchestrating gpt-4 for context-aware asset selection from objaverse, automated blender scene assembly, and dynamic skybox generation via hugging face. it bridges the gap between static text and experiential learning by synthesizing wikipedia data into narrated, spatialized historical simulations.

next.js, objaverse, huggingface api, openai api, wikipedia api, blender

learnspace

learnspace is a semantic bookmarking ecosystem that leverages snowflake and gemini to execute incremental clustering and information extraction from saved web content. the platform automates knowledge retention by converting idle links into a structured graph of interactive modules, personalized text deep-dives, and ai-generated audio summaries.

next.js, snowflake, gemini, chrome extension

parrot

parrot is a real-time ai phonetics engine that utilizes azure speech services for granular phoneme-level analysis and mediapipe facemesh for low-latency lip tracking. it gamifies pronunciation practice through an adaptive scoring system for accuracy and intonation, using gemini to generate targeted phonetic challenges that reward progress and mastery.

next.js, azure speech services, mediapipe, gemini, supabase

creditjam

creditjam is an algorithm experiment designed to test if a homemade algorithm could compare to mainstream credit card reommendation platforms without the help of generative ai. it functions as a personalized recommendation engine that analyzes spending habits and financial goals to match users with an optimized credit card portfolio (accounting for hotel and airline ecosystems). built with next.js and typescript, the system leverages a scoring logic to navigate a database of u.s. cards to make recommendations.

next.js, react, typescript, tailwind css

winsight

winsight is a computer vision project that uses a flask backend to bridge a yolov8 object detection model with a real-time web interface. the application processes live camera feeds through opencv to identify playing cards and calculate the mathematical probability of specific blackjack outcomes. it then uses an algorithmic decision engine to suggest optimal moves, like hitting or standing, based on the current deck state and statistical risk assessment.

opencv, yolov8, flask, python

design work

virginia tech entrepreneurship club website

i designed and developed this website from scratch. upgraded existing brand kit and created an asset library for the club to use.

next.js, framer motion, figma

aerostrategy website

i designed and developed aerostrategy's core platform in next.js, transitioning the site from a static landing page to a functional defi hub. the overhaul included integrating a web3 backend and implementing dune analytics dashboards to visualize live smart contract data.

next.js, figma, dune analytics

bytecode collective website

built this website for bytecode collective, a team of web3 contractors under a fun creative direction: raw, brutalist, 3d, retro. tried blending 3d elements with a lo-fi digital feel. i tried to make the site feel like a visual playground with experimental layouts and high-contrast typography.

next.js, three.js