Minjae Kang.
Frontend Developer
Introduce.
I am Minjae Kang, a frontend developer specializing in React, Next.js, React Native, Expo, TypeScript, Redux and React-query.
Taking a leave of absence from university, I ventured into startup founding and began coding to build our product from the ground up. In that journey, I discovered the immense joy of solving complex real-world problems with code and shipping products directly to users, and as a frontend developer, I now strive to turn that joy into tangible value for users.
I believe engineering is more than just drawing screens—it is about resolving business challenges and delivering tangible value to users. Drawing upon hands-on full-stack experience architecting everything from frontends to backends and cloud infrastructure, I possess a broad understanding of end-to-end system data flow. With empathetic communication and relentless execution alongside backend engineers and cross-functional partners, I strive to make a measurable impact on product and organizational growth.
🏝️Rescue the developer trapped on the island writing his resume.
Work Experience.
JustTrip - AI Integration for PDF Vouchers & Receipts.
2026.08 - 2026.09Problem Definition
Managing travel itineraries and expenses required manually typing in 10 flight fields, 6 lodging fields, and 10 expense fields. While integrating the Google Places API reduced this to 8, 5, and 8 fields respectively, manual entry remained overwhelmingly tedious. In particular, deciphering and typing details from English or local-language PDF vouchers and receipts led to frequent user frustration and input errors.
Solution
Integrated Azure Container App to process and analyze unstructured voucher PDFs and receipt photos uploaded from web and mobile clients in real time. Parsed payment timestamps, multi-currency amounts, line items, and check-in schedules using pre-trained OCR models into structured JSON, automating pre-fill and entry into frontend forms.
Results
Drastically reduced manual input fields down to just 1 action (file upload) for flights and lodging, and 3 select forms for expenses. Achieved 95% voucher parsing accuracy, reduced multi-leg itinerary registration time by 80%, and cut manual expense ledger entry effort by over 90% via automated receipt extraction.
Tech Stack
Next.js, React Native (Expo), TypeScript, Azure Container App, Azure AI Document Intelligence (OCR)
JustTrip - Product Planning & Frontend Engineering.
2026.03 - 2026.08Problem Definition
JustTrip is an AI-powered personalized travel itinerary and tracking service. Moving beyond a conventional travel app, it deeply intertwines multi-leg schedules, map routes, and real-time expense tracking, leading to frequent cross-screen state changes and duplicate API calls. Furthermore, we focused on solving the critical travel-specific issue where users cannot access flight/hotel vouchers or schedules during in-flight mode or in overseas roaming dead zones without network connectivity.
Solution
Leveraged React Query for server-state caching to eliminate duplicate API requests and centralized client UI state with Redux Toolkit. Built a two-tier offline caching pipeline tailored for travel environments: lightweight text metadata (trips, flights, accommodations, schedules, expenses) was persisted locally using Redux Persist and AsyncStorage, while multi-megabyte PDF vouchers and confirmation images bypassed standard storage limits by being pre-downloaded and cached as binary Blobs in IndexedDB.
Results
Guaranteed 100% reliable offline accessibility, enabling instant retrieval of large PDF vouchers and complete travel itineraries even in airplane mode or network-dead zones. Eliminated redundant API traffic with React Query and combined optimistic UI updates with Undo rollback interactions to successfully launch a resilient cross-platform travel service.
Tech Stack
Next.js, React Native (Expo), TypeScript, React Query, Redux Toolkit, Redux Persist, IndexedDB, AsyncStorage
RiderTown.
Co-founder / Full-Stack Developer
2022.11 - 2025.03 (2 yrs 5 mos)NonSteelKing - Non-Ferrous Metal Order Platform & Admin Dashboard.
2025.02 - 2025.03Problem Definition
The non-ferrous metal distribution and machining business involves numerous variables and complex calculations based on material types, dimensional specs (thickness, width, length, tube profiles), and machining operations. Manual quotation processes conducted over phone and messenger resulted in frequent order omissions and quote delays, with no data-centric admin system to monitor the end-to-end workflow from order receipt and unit cost estimation to machining status transitions.
Solution
Collaborated with product planners and business operators to define the quotation-order-machining lifecycle, developing an intuitive customer estimate request form and a dedicated administrative dashboard. Designed a precision form validation schema utilizing Zod and React Hook Form to dynamically validate multi-dimensional specs and processing options, precluding runtime input errors. In the admin portal, built a real-time order tracking pipeline with status-based tab filtering (Requested, Confirmed, Completed, Cancelled), unit price/lead time entry, and comprehensive audit history logging. Continuously refined user and admin UX based on feedback, introducing machining allowance emphasis, customer profile auto-population, and form preservation on page refresh.
Results
Consolidated fragmented quotation and ordering channels across phone and messengers into a unified platform, reducing manual order omissions to zero. Precluded erroneous dimensional inputs through rigorous schema validation and significantly slashed quotation turnaround lead times via real-time unit pricing and status workflow management in the admin dashboard. Furthermore, drastically curbed routine status inquiries by providing customers with a real-time order tracking timeline. Above all, actively coordinated complex manufacturing domain logic and transparently communicated engineering updates and roadblocks with non-technical business stakeholders and planners, cultivating robust cross-functional communication and agile collaboration skills.
Tech Stack
Next.js (App Router), TypeScript, MUI (Material-UI), React Hook Form, Zod, Firebase (Firestore, Auth, Storage)
RiderTown V2 - Infrastructure Migration to Firebase.
2024.09 - 2024.12Problem Definition
Following our pivot to completely sunset the maintenance booking service, maintaining the existing complex MSA architecture incurred significant fixed costs for 24/7 instances of EC2, RDS, VPC, S3, and API Gateway. Additionally, during real-time riding tracking and voice group chat, mobile devices entering the background or turning off screens caused WebSocket disconnects, resulting in lost or severed telemetry data.
Solution
We evaluated consolidating the MSA into a single monolithic server, but fixed 24/7 server running costs persisted and mobile background WebSocket loss remained unsolved. Furthermore, while the maintenance service required complex multi-table JOINs across bikes, shops, bookings, and payments, the pivoted riding service focused purely on users and crews, drastically reducing the need for an RDBMS. Consequently, we migrated entirely to a Firebase serverless architecture to eliminate fixed server costs and natively support offline/background sync. We denormalized MySQL schemas into Firestore NoSQL documents to minimize query costs, and leveraged Firebase Realtime Database for seamless background operation, utilizing onDisconnect handlers to guarantee zero data loss during background transitions or network disconnects.
Results
Reduced fixed server infrastructure costs by over 90% and eliminated server operational overhead. Guaranteed robust data consistency and synchronization across background states and unstable cellular networks during high-speed riding.
Tech Stack
GCP, Firebase (Firestore, Cloud Functions, Auth, Realtime Database), NoSQL
RiderTown V2 - Riding Service Pivot & Optimization.
2024.06 - 2024.09Problem Definition
User analytics revealed that real-time GPS riding tracking and crew location sharing drove more than 10x higher user retention than the maintenance booking feature, prompting a core strategic pivot. However, the legacy V1 approach of streaming raw coordinate objects ({latitude, longitude}) every second accumulated thousands of coordinates on long tours, wasting mobile network bandwidth and causing severe battery drain and device overheating.
Solution
Redesigned the application UI/UX from the ground up to prioritize real-time riding telemetry and interactive maps upon launch. Replaced the inefficient raw JSON coordinate streaming with Google Polyline Encoding (encodePolyline) to compress thousands of coordinates into a single compact string. Furthermore, implemented dynamic speed-based transmission intervals, a 30-second heartbeat trigger upon stopping for 15+ seconds, adaptive distance sampling (20m/40m/60m), and applied a minimum 4-second throttling to real-time path updates to maximize transmission efficiency.
Results
Cut network data payload by over 70% and drastically reduced smartphone battery drain and overheating during long-distance rides, successfully establishing the app as an essential, reliable companion that riders safely keep active throughout their journeys.
Tech Stack
Node.js, WebSocket, React Native, Geolocation API, Redux, Polyline Algorithm, Mapbox
RiderTown V1 - Group Real-Time Voice/Video Streaming Engine.
2023.11 - 2023.12Problem Definition
After introducing real-time voice chat using WebRTC, in an N-user call, each client had to handle (N-1) upload streams, requiring each user to consume 2 × (N - 1) bandwidth. As soon as 3 to 4 riders joined, this saturated mobile uplink bandwidth, resulting in severe audio stuttering and extreme device overheating.
Solution
Engineered an SFU media pipeline based on Mediasoup v3 where the server routes RTP packets without re-encoding. This locked each client’s upload stream to exactly 1 regardless of participant count, while keeping downloads at (N - 1), dramatically slashing total streams per client from 2 × (N - 1) down to N. Additionally, built a WebSocket-based voice room session management system to oversee room join/leave states and automatically reclaim media routers and socket resources upon abnormal disconnections.
Results
By fixing the per-client upload stream to 1, reduced mobile uplink bandwidth consumption by approximately 80% in a 4-person group call and stably maintained crystal-clear, zero-latency voice communication quality.
Tech Stack
Node.js, WebSocket, Mediasoup v3, Express, WebRTC
RiderTown V1 - Riding Tracking Service Development.
2023.09 - 2023.11Problem Definition
To evolve into a comprehensive motorcycle platform, we added various features—such as riding crews, community forums, and used bike marketplaces—to drive consistent user return visits. In that process, we introduced real-time ride tracking to record journeys alongside crew synchronization to view fellow riders’ logs in real time.
Solution
Calculated ride speed, distance, elevation, and top speed with precision using device Geolocation combined with useMemo compute caching. In particular, converted gyroscope sensor data into lean angles to track real-time banking angles and peak lean angles during cornering. Furthermore, streamed this personal telemetry bidirectionally via WebSockets at 1-second intervals, allowing riders to observe their crew members’ live riding states on the map.
Results
Received enthusiastic user feedback by providing motorcycle-specific metrics (banking angles) and live crew data synchronization, establishing the feature as a killer utility riders kept on during rides and substantially driving user retention and community dwell time.
Tech Stack
React Native, Node.js, WebSocket, Geolocation API, Gyroscope Sensor, Redux
RiderTown V1 - Maintenance Booking System Hardening.
2023.08 - 2023.08Problem Definition
During peak weekend and evening hours, multiple riders frequently attempted to book the same lift slot concurrently, causing race condition conflicts. In particular, during the tens of seconds while a user navigated external payment gateway (PG) checkout screens, the system failed to securely hold the slot, resulting in double bookings, payment cancellations, and refund disputes.
Solution
We initially considered WebSockets to broadcast slot selection states, but encountered a fatal mobile vulnerability: when users switched to banking or credit card apps to authenticate payment, the app entered the background, immediately severing WebSocket connections. To guarantee data integrity regardless of network state, we implemented Redis-based distributed locking. Upon entering checkout, an atomic provisional lock with a 5-minute TTL was placed on the slot, which finalized on payment confirmation or automatically rolled back if the user abandoned checkout.
Results
Eliminated 100% of double bookings and achieved 100% payment-booking transaction consistency even during high concurrency spikes, earning deep service reliability from both riders and partner repair shops.
Tech Stack
Java, Spring Boot, Redis, MySQL
RiderTown V1 - Next.js Migration.
2023.06 - 2023.08Problem Definition
Due to limited engineering resources and timeline constraints, we opted for a webview-based architecture inside React Native for rapid cross-platform delivery and maintainability. However, in our initial React SPA setup, a monolithic JS bundle exceeding 2.5MB caused 3.5+ second initial loading delays and white-screen flashes in mobile webviews, creating an unacceptably slow first impression and driving high drop-off rates.
Solution
Initially unfamiliar with Next.js, we experimented with numerous workarounds to trim bundle size: optimizing heavy packages, dynamic import code splitting, and even pre-loading upcoming webview screens in the background. However, these failed to meaningfully reduce load times while introducing severe mobile memory overhead. Through deeper technical exploration, we learned Next.js and executed a full migration to server-side rendering (SSR), effectively eliminating redundant initial bundle downloads through route-based code splitting.
Results
Slashed initial JS bundle size from 2.5MB to 550KB (approx. 78% reduction) and cut first contentful paint (FCP) time from 3.5s to 0.8s (approx. 77% faster), completely resolving webview white screens and achieving smooth, native-grade responsiveness.
Tech Stack
React, Next.js, React Native Webview
RiderTown V1 - Initial Architecture & Launch.
2022.11 - 2023.06Problem Definition
As an active rider, I personally suffered from opaque repair estimates and overcharging, and noticed frequent grievances voiced on major motorcycle community forums. I joined riding crews to survey rider sentiment and conducted in-person interviews with 100 riders at Jamsu Bridge. The survey revealed that 74% held negative perceptions of the maintenance market. Recognizing that the niche motorcycle industry was steadily expanding with strong market demand for price transparency, we set out to launch a dedicated maintenance booking platform.
Solution
Rallied a business development lead and a backend engineer to co-found a 3-person startup. Beyond self-studying and leading frontend development, I was determined to master the entire system architecture and service flow, actively learning and co-developing the backend in Java/Spring Boot. In addition, proactively systematized team development and deployment workflows by spearheading Git-flow branching strategies, Docker container-based deployment automation pipelines, and AWS (EC2, RDS, S3) infrastructure environments.
Results
Spearheaded end-to-end development spanning web and mobile app frontends, backend RESTful APIs, and AWS (EC2, RDS, S3) cloud infrastructure, successfully bringing the complete RiderTown V1 service to market. In particular, containerized services with Docker and built automated deployment pipelines to eliminate manual operational overhead and ensure release reliability, while cementing Git-flow branching and PR/code review standards across the team for agile, conflict-free delivery.
Tech Stack
React, React Native, Next.js, TypeScript, Redux, React-query, Java, Spring Boot, MySQL, Redis, AWS (EC2 · RDS · S3), Docker
AI Experience.
Document AI & Data Pipeline
Intelligent Voucher & Receipt Auto-Parsing
Data Pipeline Engine Built on Azure Container App & Azure Document Intelligence
Problem Definition
Managing travel itineraries and expenses in JustTrip initially required manual entry across 10 flight fields, 6 lodging fields, and 10 expense fields. While integrating the Google Places API reduced this to 8, 5, and 8 fields respectively, typing in numerous details while reading voucher PDFs and receipts in English or unfamiliar foreign languages was still highly tedious and prone to user error.
Solution
Engineered a real-time ingestion pipeline integrating Azure Container App to process voucher PDFs and receipt images. Using pre-trained document and receipt models, automatically extracted payment timestamps, multilingual currencies, line items, and check-in schedules into structured JSON, instantly populating itinerary schedules and expense ledgers.
Results
Reduced manual input fields down to just 1 action (file upload) for flights and hotels, and 3 select forms for expenses. Achieved 95% parsing accuracy for travel booking vouchers (slashing multi-leg itinerary registration time by 80%) and 92% OCR accuracy for multilingual receipts, cutting manual expense entry effort by over 90%.
Tech Stack
Azure Container App, Azure AI Document Intelligence
Computer Vision & Synthetic Dataset
Synthetic Dataset Pipeline & YOLOv8 Object Detection Training
50,000-Image Synthetic Dataset Generation & Multi-Region Object Detection Fine-Tuning
Problem Definition
Initially attempted manual frame capture and annotation from 3-minute hunting gameplay videos, but scaling manual labeling across vast in-game maps, dozens of monster sprites, and diverse terrain obstacles proved prohibitively expensive and unscalable.
Solution
Directly extracted map backgrounds and monster sprites from game resource (WZ) files to build an automated synthetic data generation pipeline synthesizing 50,000 randomized composites (varying positions, scales, and terrain). Fine-tuned continent-specific YOLOv8 object detection models on the generated datasets.
Results
Successfully built a real-time vision model that reliably detects diverse in-game monsters and characters across complex backgrounds without manual annotation, gaining deep practical expertise in end-to-end vision AI dataset pipelines and model training.
Tech Stack
Python, YOLOv8, PyTorch, OpenCV, Data Augmentation, Synthetic Dataset
LLM & Conversational Agents
Interactive Resume AI Assistant
Real-Time Conversational Agent Powered by RAG Knowledge Injection & Prompt Engineering
Problem Definition
Static text resumes constrain space, making it difficult for hiring managers or technical interviewers to instantly pinpoint specific competencies and project context, and fail to support dynamic, exploratory queries.
Solution
Built a structured knowledge base consolidating project troubleshooting, technical stack rationale, and engineering philosophies. Deployed system prompt engineering and dynamic context injection to create a conversational assistant that accurately surfaces grounded evidence and experiences tailored to the user’s intent.
Results
Semantically parsed visitor inquiries (tech adoption rationale, debugging narratives) to present relevant project links and architectural reasoning in real time, transforming a one-way resume into an interactive, exploratory dialogue experience.
Tech Stack
Gemini API, RAG Context Injection, Prompt Engineering
LLM & Conversational Agents
Vita - Personalized AI Supplement Recommendation Engine
Health Survey Structuring & Real-Time Nutrient Analysis via LLM
Problem Definition
Faced with overwhelming supplement claims and aggressive marketing, ordinary individuals struggle to identify which nutrients align with their personal physical condition and lifestyle, and cannot easily verify potential overconsumption risks, drug interactions, or adverse side effects.
Solution
Structured user physical metrics (BMI) and lifestyle questionnaire data to design a nutrient scoring model and exclusion ruleset. Combined this with LLM prompt engineering to build a real-time pipeline that analyzes supplement compatibility (interactions) and generates customized nutrient recommendations.
Results
Generated personalized supplement analysis reports stably within 10 seconds using LLMs, and built daily intake checklist calendars and reminder routines, successfully shipping the service across web and cross-platform mobile apps.
Tech Stack
OpenAI API, Prompt Engineering
LLM & Conversational Agents
HappyWhale - Emotion-Adaptive Daily AI Notification
Real-Time Empathetic & Motivational Messages Tailored to User Emotional States
Problem Definition
Unlike typical mobile push notifications limited to cold factual information, I wanted to leverage generative AI to deliver genuine comfort and uplifting motivation tailored to the user’s nuanced emotional state and daily context.
Solution
Designed empathetic prompt templates and tuned generation parameters to avoid clichéd responses. Built an emotional care AI system that generates and schedules real-time LLM messages matched to the user’s selected emotional state and delivery time preferences.
Results
Delivered personalized LLM emotional messages with high uptime, providing positive emotional care experiences to over 1,300 active users.
Tech Stack
OpenAI API, Prompt Engineering
Other Experience.
RiderTown.
Co-founder & Field Sales / Partnerships
2022.11 - 2024.03Repair Shop Onboarding & Field Sales.
2023.06 - 2023.08Problem Definition
RiderTown V1 was specifically tailored for motorcycle maintenance quotes and bookings, making repair shop recruitment essential. With zero initial capital or brand recognition, we wrestled with how to convince and onboard repair shops onto an unproven platform.
Solution
Immediately after completing development, cold-called repair shops to introduce the service and secure in-person meeting appointments. Printed physical business proposals and travelled nationwide alongside my co-founder to meet shop owners on-site, building genuine trust and securing partnerships through persistent, direct communication.
Results
Successfully onboarded 41 high-performance motorcycle repair shops nationwide (excluding Jeju). Identified real-world operational needs through direct conversations with shop owners and continually refined the service accordingly. Through this experience, I deeply internalized how to communicate effectively with cross-functional stakeholders and on-site professionals as a developer.
Early Rider User Acquisition.
2023.06 - 2023.08Problem Definition
With absolutely zero initial marketing budget, acquiring genuine motorcycle riders was vital. Without active end users, our partner repair shops would see no real value in remaining on the platform.
Solution
Printed app download QR codes onto business cards and visited high-density rider hotspots such as Jamsu Bridge parking lots and rider cafés, introducing the service 1-on-1. Formed mutual partnerships with rider cafés by placing promotional posters on-site in exchange for featuring the cafés on our platform.
Results
Acquired an active initial user base with minimal marketing spend, gathered direct real-world market feedback, and established the critical traction baseline required to secure major government grants including the Pre-Startup Package and Youth Startup Academy.
Award.
Korea SMEs and Startups Agency (KOSME).
2024.03 - 2024.11Youth Startup Academy (Startup Success Package).
Motorcycle Consumables Lifecycle & Maintenance Platform via Riding DB [100M KRW Scope]
Korea Institute of Startup & Entrepreneurship Development (KISED).
2023.05 - 2023.12Pre-Startup Package.
Motorcycle Repair Quotation & Community Platform [40M KRW Scope]
Skills.
Overall.
Prefer clean user interfaces, polished visual hierarchy, and fluid micro-animations.
Prioritize choosing the most appropriate and pragmatic technology over simply chasing trends, while continuously learning the modern frontend ecosystem.
Proactively research and dive into unfamiliar domains when necessary to deliver optimal business outcomes.
Communication.
Proactively voice ideas, architectural proposals, and product improvements.
When encountering ambiguities or blockers, pinpoint the exact unknowns and formulate precise, constructive questions.
Always prioritize active listening and empathy before asserting my own viewpoint.
JavaScript / TypeScript.
Proficient in JavaScript (ES6+) and TypeScript language ecosystems and runtimes.
Understand module systems including CommonJS and ES Modules, leveraging appropriate tooling.
Establish strict data contracts between modules and components via type definitions and interfaces, maximizing developer velocity and code maintainability.
React / Next.js.
Skilled in React Hooks and custom hooks to modularize reusable business logic and design clean, composable components.
Proficient in designing optimal rendering strategies (SSR, SSG, ISR) and SEO structures in Next.js tailored to product requirements.
Experienced in designing robust server-state caching and global state architectures using Redux and React Query.
Experienced in performance optimization through Core Web Vitals profiling, dynamic imports, and route-based code splitting.
React Native / Expo.
Capable of delivering cross-platform iOS and Android mobile apps from concept to production app store release using React Native and Expo.
Experienced in integrating native device capabilities such as real-time GPS tracking, background tasks, camera, and push notifications.
Design responsive layouts and smooth gesture animations tailored to platform-specific UI/UX guidelines.
Skilled in automated build/deploy pipelines using CodePush and EAS (Expo Application Services), with rich experience navigating App Store and Google Play reviews.
Backend / Database.
Designed and implemented scalable RESTful APIs using Java Spring Boot and Node.js.
Experience designing relational schemas and query optimization in MySQL, as well as scalable NoSQL document modeling in Cloud Firestore.
Applied distributed locking with Redis for concurrency control and data integrity, such as atomic booking slot reservations.
Experienced in event-driven asynchronous messaging and system decoupling using RabbitMQ.
DevOps.
Standardize, isolate, and operate development and production environments using Docker containerization.
Architect and operate stable, scalable cloud infrastructures leveraging AWS EC2, RDS, S3, and CloudFront.
Construct automated CI/CD deployment pipelines using GitHub Actions.
Education.
Kwangwoon University.
2018.03 - 2026.08B.S. in Electronic Engineering
Sanbon High School.
2013.03 - 2016.02Natural Sciences Track
Monthly Minjae.
Motivation
Whenever my mother filmed badminton tournament videos, the large file sizes made it impossible to send them via messaging apps or email, so she would ask me to edit or compress them. Noticing she found complex cloud storage drives confusing, I built this service to help her out (and certainly not because I found her requests annoying!).
Solution
Built a lightweight web utility allowing drag-and-drop uploads and instant one-click link sharing without requiring complex cloud drive accounts. Implemented resumable chunk uploads to prevent network disconnects and engineered in-app browser bypass schemes (KakaoTalk / Naver) to ensure smooth full-resolution video downloads on mobile.
Tech Stack
React, Next.js, Firebase (Storage · Firestore)
Motivation
Watching my younger sibling prepare for college entrance exams while struggling with and disliking English, I wanted to provide positive motivation rather than forcing study habits, drawing upon my own experience maintaining top marks.
Solution
Curated 5,000 essential CSAT vocabulary words and built an automated data pipeline using the Gemini API to generate exam-tailored example sentences and translations. Combined intuitive swipe flashcards, Ebbinghaus forgetting curve-based spaced repetition review, and native TTS pronunciation to deliver a tailored free mobile vocab study app.
Tech Stack
React Native, Expo, TypeScript, Redux Toolkit, Expo Speech, Gemini API, Firebase
Motivation
Developed to automate repetitive in-game hunting for my sibling while diving deep into computer vision object detection and vision AI dataset training pipelines.
Solution
Initially captured 200 screenshots from a 3-minute hunting clip for manual labeling. To scale across all maps and monsters, extracted sprite assets from game WZ resource files to synthesize a 50,000-image dataset and trained YOLOv8 continuously for 3 weeks—until my 10-year-old computer finally broke down, ending both my training and my sibling’s gaming session.
Tech Stack
Python, YOLOv8, PyTorch, OpenCV
Motivation
After catching colds every seasonal transition and starting Vitamin C megadosing, I grew exhausted navigating hundreds of complicated supplement claims and wanted to build a straightforward solution myself.
Solution
Architected a scoring engine and contraindication exclusion ruleset based on health questionnaire responses, built real-time supplement interaction (compatibility) analysis, and launched an intake checklist calendar and reminder routine across web and mobile stores.
Tech Stack
React Native, Expo, React, Next.js, TypeScript, Redux Toolkit, Notifee, Firebase, OpenAI API
Motivation
Conceived as a global casual game allowing worldwide tournament matchups with rewards. However, with university coursework and concurrent projects becoming overwhelming, I scoped down and shipped a polished standalone game to avoid indefinitely shelving the release.
Solution
Leveraged Firebase Realtime Database runTransaction to prevent race conditions during concurrent bets, utilized onDisconnect to handle forfeits and clean up abandoned rooms, and released on both app stores with landscape arcade UI, haptic/audio feedback, 7-language localization, and in-app purchases.
Tech Stack
React Native, Expo, TypeScript, Firebase (Realtime Database · Firestore), Expo Audio & Haptics, Expo IAP
Motivation
Every week at amateur soccer matches, club managers awkwardly typed out rosters and tactical formations in chat rooms. I created this tool so anyone could visually arrange and share tactics with ease.
Solution
Built relative-coordinate (0~100%) drag-and-drop player positioning via Framer Motion across all screen sizes, implemented seamless switching between tactical formations (4-3-3, 4-4-2), integrated custom squad mapping, and added real-time Firestore sync with one-click link sharing.
Tech Stack
React, Next.js, TypeScript, Framer Motion, MUI, Firebase (Cloud Firestore)
Motivation
During a challenging personal period, receiving warm notification messages from Toss inspired me to build notifications that comfort and encourage the user rather than merely delivering dry information.
Solution
Engineered a Cloud Run AI message pipeline, wrote a native iOS Notification Service Extension in Swift to render messenger-style profile avatar rich pushes, and integrated Notifee and FCM for precision scheduling.
Tech Stack
React Native, Expo, TypeScript, Swift (iOS Notification Extension), Notifee, Firebase (FCM · Firestore), OpenAI AI
Motivation
After recommending Attack on Titan to my sister and rewatching it together—truly the greatest masterpiece—I felt inspired. Combined with fond memories of traveling in Japan with friends and a desire to communicate with my aunt and uncle living there, I set out to self-study beginner Japanese in an enjoyable, interactive way.
Solution
Engineered an interactive SVG stroke-order writing guide for Hiragana and Katakana, native TTS audio pronunciation, gesture-driven flashcard swiping, and a 4-choice quiz loop to reinforce foundational character memorization.
Tech Stack
React Native, Expo, TypeScript, Redux Toolkit, React Native SVG, Expo Speech
Motivation
After a quarrel with my girlfriend, to break the awkward tension playfully rather than sending a blunt text, I devised a curiosity-driven communication tool that conceals photos and secret messages behind a scratch-off mosaic blur.
Solution
Built on HTML5 Canvas and Konva to capture user drawing paths (coordinates, timestamps, stroke widths) normalized as aspect-ratio-independent relative coordinates. Generated canvas mosaic blurs and engineered a real-time handwriting playback animation that triggers once 80% or more of the mosaic mask is scratched off.
Tech Stack
React, Next.js, TypeScript, Konva (HTML5 Canvas), Firebase (Firestore · Storage)