LEO Satellite Routing with Deep RL / h-DQN

• Problem: Energy-aware ground station selection (10 stations) for LEO downlink with AoI (Age of Information) minimization under queue capacity + distance-based communication energy constraints.
• Method: Built a DQN baseline that trains stably (≈10k episodes) in a TLE-driven OCO-2 orbit/visibility simulation (step-level logging: reward/AoI/energy/queue).
• Evaluation: Compared against visibility-aware Random and distance-greedy baselines; reporting return + AoI/energy/queue dispersion (mean±std over N eval episodes).
• Next: Extending to hierarchical DQN (h-DQN) to decouple high-level objective selection vs low-level station choice for more traceable decisions and faster fault isolation.
• Paper: KICS 2025 (oral) · DBpia

Multimodal & Hybrid Embeddings for RAG

• Text-Image Alignment: Developed a Multimodal RAG pipeline using CLIP-based joint embeddings to bridge semantic gaps between disparate data types.
• Hybrid Retrieval Strategy: Combined dense (vector) and sparse embeddings to improve semantic comprehension by ~15%; optimized for production via Docker & K8s.
• Research Impact: Evaluated performance across domain-specific datasets, recognized at KSC 2024 (Best Paper) and KICS 2025.

Production Agent & ML Systems

• Agentic Workflow: Built MCP-based tool orchestration and ReAct(DSPy) agents with 96%(REGAS) step-level observability (@ Rock15 Inc).
• Advanced RAG: Developed GraphRAG pipelines involving KGE model training and structured document indexing (@ Rock15 Inc).
• MLOps & Rec System: Deployed hybrid XGBoost/NN recommendation models with automated CI/CD pipelines (@ Q-Click Inc).

• Interpretable & reliable reinforcement learning (DQN, multi-agent, sequential decisions)
• ML systems: observability vs efficiency, MLOps, Graph RAG, agent architectures
• AI + Systems: building practical systems that bridge theory and production under real constraints

AWS GenAI (2nd), KSC Best Paper (3rd), KICS GenAI (3rd), KCCI AI Startup (2nd),
Shinhan Big Data Hackathon (2nd), COKOTHON Inter-University Hackathon (Grand Prize, Wiggle)

More details & certificate photos: LinkedIn

• Rock15 Inc. — AI Research Engineering Intern (San Jose, CA)
• Distributed Platforms & Systems Lab — Undergraduate Researcher (Satellite RL)
• Data Intelligence Lab — Undergraduate Researcher (Multimodal RAG)
• Dacos ML Club — Member & PyTorch training mentor
• Q-click Inc. — MLOps Engineering Intern (Seoul, KR)
• UMC @ SMWU — 5th Member (Seoul-wide developer community)
• Solux @ SMWU — 28th Member (Application development club)
• APPS @ SMWU — 10th Member (Application development club)

• Email: chaestaryinhw12@gmail.com
• LinkedIn: linkedin.com/in/chloe122