Gijin Kim
Backend Developer & AI Engineer
A passionate developer who enjoys simplifying complex problems and solving them with IT technology. From handling large-scale traffic to developing AI services, I focus on creating practical value through various technical challenges.
About
I'm Gijin Kim, a developer who finds fulfillment and happiness in solving complex real-world problems with IT. I focus on building software that delivers value to people, from handling large-scale traffic to developing RAG-based intelligent services. Recently, to expand my ability to solve more problems, I have been studying natural language processing and pursuing related research.
Education
Experience
Responsible for technical leadership and team management including scrum facilitation, code reviews, API design, and RDB/NoSQL optimization across projects.
- Designed and developed chat server architecture handling 400K DAU traffic
- Verified message processing stability through Artillery-based load testing
- Developed read receipts, 1-on-1/group/channel chat (5M+ simultaneous delivery)
- Node.js, Nest.JS API design with Socket.io real-time communication
- Implemented multi-tenant architecture for system scalability
- MySQL, DynamoDB, Redis design and optimization for performance
- Designed and developed backend systems for North America and Europe
- Built stable architecture considering network latency and data consistency
- Maintained features in AWS cloud (EC2, RDS, S3, SQS, Lambda)
Projects
AI-Powered Slack Q&A Bot
RAG-based Slack bot providing accurate, evidence-based answers for internal team knowledge (GitHub, Notion)
- Question classification and routing workflow using LangGraph
- Vector search with Gemini Embedding Model and Pinecone
- Periodic GitHub/Notion change detection and embedding updates
- Multi-tenancy DB schema and logic design
RAG Pipeline Optimization
Quantitative analysis of TopK, Chunk Size, Overlap parameters' impact on response quality and cost
- 27 parameter combinations × 21 questions × 3 iterations
- RAGAS framework for quality metrics measurement
- Derived optimal point (TopK=3, Chunk=512, Overlap=15%)
- Answer Correctness +17%, Token usage -57%
AI/ML Study Blog
Structured blog covering Transformer, CNN, ViT, U-Net, and Pattern Recognition fundamentals
- Transformer & Attention mechanism documentation
- CNN series: LeNet, AlexNet, VGG, ResNet, MobileNet
- Pattern Recognition: MLE, SVM, Gradient Descent
Food Image Calorie Analysis
AI API server analyzing calories and nutritional information from food photos using fine-tuned YOLO
- YOLOv11 fine-tuned with 90 food classes (mAP 0.85)
- FastAPI real-time image processing pipeline
- AWS cloud-based scalable server infrastructure
AI Clothing Recycling Platform
Service automating clothing recycling sorting with YOLO models
- 15-category classification and defect detection API
- TorchScript conversion for inference optimization
- Docker Hub + AWS Elastic Beanstalk auto-deployment
Gym Attendance Motivation App
Fitness web/app with gym attendance and social features
- Gym image verification using OpenRouter AI
- Real-time ranking and level system with Redis
- User motivation through social features
Skills
Programming
AI / ML
Backend & DevOps
Research Interests
RAG & Domain-Specific LLM
Based on hands-on experience building RAG-based bots, I am interested in research that leverages external information sources to enhance LLM reliability and reduce hallucinations. Currently exploring efficient search, indexing, and generation strategies for domain-specific data.
LLM Serving & Inference Optimization
Recognizing bottlenecks and throughput limitations in LLM inference as key challenges, I am exploring optimization problems in inference pipelines and serving architectures.