Built an end-to-end RAG pipeline using Python, FastAPI, LangChain, and ChromaDB to automate document QA for an enterprise invoice management system; improved retrieval quality by tuning chunk size/overlap and integrated enterprise data via REST APIs. Configured and deployed an OpenClaw gateway on a Mac Mini server to route requests between cloud APIs (Azure OpenAI, Gemini) and local open-source models (Gemma via Ollama) for offline capability and cost/speed optimization. Developed a real-time face recognition attendance tool (InsightFace/ONNXRuntime) using event-driven tracking logic; collaborated with the "antigravity" team to sync metadata and refactor core logic while maintaining interfaces. Built OMR auto-grading pipelines (Python, OpenCV) handling skew/lighting; developed STT modules to automate meeting minutes extraction. Packaged AI desktop utilities into executables using PyInstaller and provided handover documentation.

Researched GAN architecture and image synthesis principles. Implemented and trained generator/discriminator models to create and evaluate artificial images. Optimized parameters to improve visual quality and accuracy; presented findings to the university evaluation panel.