I am interested in structurally analyzing the bottlenecks that arise when multimodal AI systems are deployed in physical environments, and in addressing them from algorithmic, system-level, and hardware perspectives.
PDF λ³΄κ³ μλ₯Ό ν¬νΈν΄λ¦¬μ€μ© λ§ν¬λ€μ΄μΌλ‘ μ¬μμ±νκ³ Notionμ ν λ²μ μ μ₯νλ μμ΄μ νΈ. Upstage Document Parse + Solar LLM, Streamlit UI, Notion API μ°λ.
Tree-based compressor reducing 12 one-bit inputs to 4-bit output using Full Adders and Half Adders. FreePDK45 technology with KLayout/ngspice. DRC PASSED, hierarchical LVS analyzed with improvement plan.
π 1st Place Winner! MCP and LangGraph-based AI agent orchestration system for centralized personal work information management with automated daily briefing.
Fine-tuned Segment Anything Model (SAM) for dental X-ray segmentation using LoRA-based tuning strategies. Joint first authorship with Yoonhee Uhm.
Built hierarchical scheme clustering 100 classes into 20 superclasses with L_class + Ξ±Β·L_superclass. Applied two-stage retraining with class-balanced sampling and LDAM. Results: Overall Top-1 48.8% β 75.7%, Tail 36.6% β 70.3%.
π μ°μμ μμ (43κ° ν μ€ 2λ±)! MFC λ΄ λλ‘ μλ μ λΉ μμ€ν PoC ꡬν. Elephant Roboticsμ MyAGV2023μ Cobot320μ νμ©ν Smart Factory λ° CMMS ꡬμΆ.
3D-CNN + LSTM model for 6-frame sequences; ablations on dropout, LR schedulers, test-time normalization; MAE = 1.82.