Nathaniel's Projects

SeoulFlow

Real-time traffic congestion monitoring system using live Seoul CCTV footage. Detects and tracks vehicles with YOLOv8 and Kalman Filter, classifies congestion levels, and visualizes traffic density across key Seoul intersections.Tech Stack: Python, YOLOv8, OpenCV, Kalman Filter, FastAPI, CUDA, Chrome Extensions (Manifest V3)

EPC Procurement Intelligence Pipeline

Production-grade data engineering pipeline for Oil & Gas procurement intelligence. Medallion Architecture (Bronze → Silver → Gold) tracking supplier risk, trade trends, and material category performance across 7 countries and 92 materials.Tech Stack: Python, PySpark, Supabase PostgreSQL, Streamlit, Pandas, Plotly, Parquet, SQLAlchemyData Source: UN Comtrade · HS Chapter 84 (Industrial Machinery)

3D Reconstruction from Multiple Images

Graduation project where I implemented and compared NeRF, Vanilla 3DGS, Scaffold-GS, and Feature-GS on self-collected multi-view datasets, modified the 3DGS CUDA rasterizer to resolve OOM errors on an RTX 2070 (8GB VRAM), enabling full training under hardware constraints.
Integrated SAM embeddings into Feature-GS for 3D object segmentation and interactive manipulation via Open3D.Tech Stack: Python, PyTorch, COLMAP, NerfStudio, Open3D, SAM, MeshLab, CUDA 11.8

StartupAI

Multi-agent AI platform where LLM agents (CEO, CTO, CFO, COO) collaborate in a roundtable to auto-generate startup pitch decks. Built the LangGraph orchestration pipeline and initial MVP independently.Tech Stack: Python, LangGraph, OpenAI GPT-4o, RAG Pipeline, Streamlit

Sign Language Detection

Real-time hand landmark detection using MediaPipe and a Random Forest classifier for sign language gesture recognition. Built a custom dataset and used OpenCV for live video stream preprocessing.Tech Stack: Python, OpenCV, MediaPipe, Scikit-Learn

PetMeeting

Flutter app for connecting pet owners. Spearheaded project planning and built the frontend.Tech Stack: Flutter, Dart