Builds

Tells founders which investors are most likely to say yes, why they might pass, and the warmest way in, so a first outreach list is focused instead of a generic database export.

Ranks investors by stage, thesis fit, portfolio conflicts, recent activity, and warm-intro paths. Grounded in the fundraising patterns I saw across hundreds of pitches at b2venture and 212.

A private context and routing layer for working with agents across projects. It carries project memory, source material, agent skills, routing rules, and review loops, so agents act with continuity instead of starting from zero each time.

It keeps the source trail, so agents work from a project's real history instead of me re-explaining it each time.

Compliance QA for German vehicle expert reports: AI reads the messy document, traceable rules make the call. It was built to dodge the classic AI demo trap, keyword matching that looks right on the examples but breaks on reports it has not seen.

Built in 24 hours with my team at the HackNation hackathon at IPAI in Heilbronn. The strongest part was moving from brittle pattern matching to evidence-backed checks reviewers can trace.

An adaptive study app for Boat Licence B / skipper qualification prep, built from my own exam loop, now being sold. It turns the question catalog into focused practice, weak-spot repetition, and exam simulation, so learners train against what they actually miss instead of rereading everything.

Search German parliamentary activity by asking questions across sessions, topics, and policymakers, instead of clicking through scattered official portals.

Turns user feedback into quests and rewards, so companies get sharper product signal and people actually want to answer.

Tests reader-shaped stories: continuing, remixing, and reshaping a narrative instead of only consuming it.

Where I test assumptions fast across social algorithms, dating, govtech, storage, utility apps, and agent workflows, and find where AI is actually doing the work.