shuxton

Worked on a pilot project funded by BMBF to develop a cost-effective, non-invasive alternative to fMRI scans for detecting neurological disorders through dysphagia analysis. Gathered requirements from doctors, conducted literature review, experimented with models (UNet, MedSam2), and defined labeling requirements. Designed a rule-based classifier and bolus tracker using classical CV techniques. Studied emergent behavior in multi-agent systems as an NLP task with PPO-based reinforcement learning in MuJoCo. Initiated SLAMCOM study (explorer–mapper agents) and built a library for simplifying environment creation (microcosm.ai).

Worked closely with product experts and mechanical engineers to develop a design and costing tool for a manufacturing company, by collaborating with global teams in Germany, Switzerland, and China. Integrated 15+ liquid distributor products, migrated legacy Python code, added solid works integration. Developed a multi-tenant tax-compliant system for Saudi Arabia with ZATCA-compliant E-invoicing, FTP integration, invoice signing, and validation modules. Created prototype RAG-based chatbots for answering database/manual queries and built a computer vision tool to automatically redact sensitive data.

Designed a reconciliation tool for Kotak Mahindra Bank, reducing IPO data processing time from 15 days to 10 minutes. Built a mobile app, My Finance Wellness, with interactive financial literacy tools and calculators. Developed services to bridge modern E-Invoicing applications with legacy AS400 systems.