Lab
Research & prototypes
Applied AI and systems research — computer vision, NLP, generative models, and data extraction. Some shipped to production, others remain focused prototypes that informed later client work.
Furniture Valuation CV Pipeline
shippedProduction computer vision system for image-based asset evaluation: valuable-vs-undesirable classification, resale value prediction, duplicate-image scam detection, CSV bulk processing at 300+ listings/min, and containerized API deployment for backend integration.
Domain: Computer Vision / ML
Interior Scene Segmentation Pipeline
shippedTrained and refined Mask2Former/SAM segmentation models for real estate interior photos. Built the full pipeline: dataset merging, COCO annotation generation, train/val splitting, SAM3 training patches, Hydra configs, and checkpoint management. Focused on tighter boundaries, edge refinement, and production-quality mask output on RTX 3090.
Domain: Computer Vision / ML
YOLO Object Detection API
shippedTrained a custom YOLO model with Ultralytics, exported to ONNX, and deployed as a real-time inference API via FastAPI + Docker. Explored M1 macOS training environment setup and production model serving patterns.
Domain: Computer Vision
Artwork-on-Wall AI Preview
explorationR&D exploration for realistic AI-powered image placement on walls. Tested ControlNet, MiDaS depth estimation, and Segment Anything for room-aware compositing. Built prototypes in Gradio and Streamlit.
Domain: Generative AI
Political Quote Extraction System
shippedShipped daily automated pipeline: scrape news sites → extract politician quotes via GPT → entity-match to Strapi database → deduplication → admin scraping form → review interface. Built scraping, extraction, and storage layers end to end.
Domain: NLP / Automation
Private LLM / RAG Pipeline
prototypeBuilt a secure retrieval-augmented generation system with document ingestion, vector store indexing, auth-aware query APIs, and structured JSON outputs for enterprise use cases.
Domain: AI Infrastructure
AI Market Insights Dashboard
shippedCombined classical data processing (Pandas, scikit-learn) with GPT narrative generation to produce analyst-quality market insight reports with interactive React visualizations.
Domain: Data + AI
Buku.app Book OCR Pipeline
prototypeAdvanced extraction pipeline: authenticated login → page-by-page capture → handle PDF/canvas/XHTML rendering → OCR fallback → structured text + metadata output.
Domain: Data Extraction