I build AI products inside large organizations, where the model is rarely the hard part.
What decides the outcome is the gap between a working demo and a system people depend on. Forward-deployed work means owning that gap — and it's the work I'm looking for.
Enterprise applications I built at Mercedes-Benz as the primary developer. The business context and requirements are shaped with the team. These systems run internally, so only the concept is shared here — not the code, screenshots, or data.
An enterprise AI transformation platform for any process, in any function. From Unstructured Documents to Organizational Change. Parses source documents into a verified BPMN map, designs the AI-native future state, and issues the implementation plan and workforce impact. Every score and FTE figure carries its source, method, and approver. A live pilot with its first users, now being engineered toward production. |
A full-lifecycle talent acquisition platform. Tracked and Supported at Every Stage. Consolidates job descriptions, resumes, interview materials, and transcripts into a single candidate record: competency models, candidate comparison, generated interview questions, AI interviews, and AI-assisted scoring. On the roadmap: an AI recruiter that sources candidates rather than waiting for applications. It supports the hiring decision; it does not make it. |
Built on nights and weekends, for no reason except that I wanted them to exist. I hope you enjoy them. Fork one, break it, tell me I got something wrong — or just say hi. And if you're building something open that looks like fun, count me in.
An agent skill that turns a sentence into a finished 3D print. From an Idea to a Physical Object. Finds or generates the model, checks printability, repairs the mesh, previews the result. You review and start the print in Bambu Studio; the agent only watches. Featured by Bambu Lab in their 2026 Skill Showcase. MIT licensed, and it runs inside Claude Code, Codex, Cursor, and 60-plus other agents. |
An LSTM–Transformer ensemble estimating whether an A-share stock closes up tomorrow. Finding Signal in Financial Time Series. Roughly 3 million sequences from 5,800 stocks, with market, sector, margin, and northbound-flow features. Ten models train on walk-forward splits, and permutation importance sorts the useful features from the harmful ones. A modeling and evaluation project, not a claim of investment performance. |
