Active Research
Docling & Multimodal Parsing
R&D
Researching the efficacy of IBM's Docling for converting complex enterprise PDFs into RAG-ready vector embeddings. We are benchmarking 'Local-First' parsing against cloud APIs to minimize latency and improve data privacy in finance and healthcare verticals.
Browser-Use Orchestration
R&D
Developing a 'Strategy/Spider' protocol to decouple AI reasoning from browser automation execution. This research aims to solve the 'Token Cost' problem in agentic workflows by using cheap, deterministic scripts for execution after high-intelligence agents have mapped the path.
Vision-Based Augmentation (MediaPipe)
Active
Researching the constraints of browser-based hand tracking using MediaPipe and React Three Fiber. Our 'poc-camera-ml' project explores coordinate mapping, physics-based interaction (Rapier), and latency-compensation strategies on low-power consumer hardware.
Open Source Contributions
Community
- TediousJS: Contributed Azure Active Directory authentication support to the node-mssql driver, enabling secure cloud connectivity for thousands of developers.
- ShareDis: A plugin for sharing links on IRC/Matrix protocols without link previews.
- DOSBox Staging Replacer: A C++ utility for GOG Galaxy users to swap default emulators. Demonstrating our love for low-level systems and retro gaming.
