· Project Update
What I have been building since the summer
Two public market tools, a Swiss German speech benchmark, an MCP server, a robotics simulator, and a few open-source evaluation projects.
The last few months were busy on the building side. Here is a short overview of what is new, with links to the project pages.
Live products
- AIUseCaseHub is public and now tracks more than 4,000 enterprise AI deployments, each linked to a public source.
- Rankoon is the public version of my XGBoost crypto research: live signals, daily rankings, paper portfolios, and published hit rates.
- PerpXray is a read-only analytics tool for Hyperliquid traders. Paste an address, get an honest breakdown of how that account trades.
Open source
- Swiss German Voice Benchmark compares voice models on eight Swiss German dialects.
- AIUseCaseHub MCP lets agents search the AIUseCaseHub record through four MCP tools.
- AI Web Scraping Pipeline is the agentic discovery, extraction, and review pipeline behind AIUseCaseHub.
- SWE-RL Forge Lite turns real pull requests into verifiable tasks for coding agents.
- RFP Agent on Microsoft Foundry shows grounded, observable multi-agent work with human approval.
In progress
- Milo is a robotics simulator where I study embodied agents, navigation, and the cost of supervision.
The common thread is evaluation. Almost every project includes a way to measure whether it works, because that is where I learn the most.
project-update · open-source · agents · evaluation