Annabel Hyunchin Lee
From perception to production, for autonomous driving.
Senior in AI Engineering · Computer Vision · Autonomous Driving · Backend
About
Dual citizen of the United States and South Korea
I'm growth-oriented: I learn by joining communities and collaborating.
I debug like a detective: gather clues, form a hypothesis, test it, even when the cause sits outside my own part of the system.
Education
Sookmyung Women's University
B.S. in Artificial Intelligence Engineering · Mar 2023 – Aug 2027 (expected)
Northern Kentucky University (NKU)
Exchange student · Aug – Dec 2025
Certificates
TOEFL 103 / 120 · OPIc AL
Skills
- Python
- C
- C++
- Java
- Spring
- Computer Vision
Projects
Autonomy & robotics
FROZEN
Camera–LiDAR perception for a MORAI-simulator driving competition (7-person team; results due late October).
- Trained YOLOv11 on self-collected data (6 classes) and fused it with LiDAR clusters to classify obstacles.
- Fixed static/dynamic confusion with ego-motion filtering, and pedestrian false positives with threshold tuning and background images.
- On the perception team, but co-built the full pipeline, so I trace failures across planning, control, and localization.
- Python
- YOLOv11
- CLRNet
- LiDAR
- DBSCAN
- MORAI
Mandro EMG Armband
Solo internship project: an EMG armband that learns a user's wrist gestures, trained entirely on the phone.
- Six gestures from forearm EMG on an ESP32-S3; on-device scikit-learn training replaced a server and TensorFlow, at 5–12× the speed.
- Only the ~52 KB of weights go to the armband over BLE, with a checksum and safe swap, so retraining needs no USB reflash.
- Tried dynamic motions; accuracy dropped, so I kept static gestures.
- Kotlin
- Android
- Python
- scikit-learn
- BLE
- ESP32-S3
Private repository: this was paid work for a company, so the code and demo aren't public.
findGuma
A P2P trading-match protocol on LoRa boards, built as two finite state machines (3-person capstone).
- Implemented the coordinator's state machine, which pairs buyers and sellers.
- Fixed duplicate transmissions with a link-layer queue and extra states that confirm sends in order.
- All state transitions ran end to end on LoRa boards.
- C++
- LoRa
- mbed
- FSM
Backend
troublelog
AI mock interviews that test whether you understand your own project (4-person team; I built the backend and infrastructure).
- Java 21 and Spring Boot backend on Azure, with GitHub Actions CI/CD.
- Azure OpenAI generates three interview questions per project and feedback on each answer.
- Responsible-AI layer: audit logs for AI calls and sensitive-data detection that blocks report generation.
- Java
- Spring Boot
- Azure
- Azure OpenAI
- Azure SQL
- GitHub Actions
let'sIT
A platform for job-seeking developers to form project teams and get an AI-made portfolio (6-person team; my first project).
- Built authentication and profiles: Kakao OAuth2 login, JWT issuance and filtering, profile CRUD with images.
- Won the Excellence Award (3rd place) in SOLUX's internal evaluation; deployed on AWS.
- The early code got messy, so the team later regrouped to refactor it together.
- Java
- Spring
- OAuth2
- JWT
- AWS
DotheG
A gamified eco-activity app: log green habits and collect characters (backend team of four).
- Built authentication and profiles: signup and login, Naver OAuth, JWT with refresh tokens, account deletion.
- The Spring Boot REST API covers quizzes, activity verification, a pedometer, and a character gacha.
- Presented in a live demo; won the Excellence Award (3rd place) in SOLUX's internal evaluation.
- Java
- Spring Boot
- JWT
- OAuth2
Experience
Internships
Goorm
Full-stack intern, 1 month
Mandro
EMG data-visualization Android app, BLE firmware modification, PC app redesign
Exchange
Northern Kentucky University (NKU)
Aug–Dec 2025 · incl. Fidelity-sponsored hackathon