I’m a longtime community builder and e-participation expert.
Founder, Web Monday. Director, Center for Applied Community Engagement (CACE). Founder, Intellitics, Inc.
Based in sunny San José, CA. Wannabe trailrunner.
Community Lead, IBM Research 2023 – 2025
Community lead for the AI Alliance, the open-source AI consortium spanning models, data, evaluation and safety. Represented it at 15+ international conferences including NeurIPS and FOSDEM, ran dozens of meetups end to end across North America and Europe. AI-assisted prototyping and web development, community ops automation.
Community Lead, IBM Research 2022 – 2025
Helped launch and build OSSci, a NumFOCUS initiative sponsored by IBM connecting open-source research software to the researchers who depend on it. Program design, global ecosystem engagement.
Community Program Manager, IBM 2019 – 2022
Community specialist on the IBM Data Science Community team, running day-to-day operations through an early growth phase. Helped take it past 25,000 members and built the global meetup outreach program.
I’m not a web developer or software engineer, so being handed these new AI superpowers has been nothing short of amazing. Over the past year I’ve shipped a range of (social) web apps using AI-assisted tools (Lovable, Claude Code, among others), covering a broad range of use cases.
Every project ships against the same baseline:
Which is where it gets interesting. Each project is run by its own AI agent, working from its own written charter. They share a versioned library of standards — how a database schema stays honest when three parties can change it, what a security baseline actually requires, how to review code an AI wrote. When one project discovers a rule is wrong, it files the finding, the library is amended, and the correction reaches every other project.
A couple of projects use AI agents for automated web search and data processing.
A few highlights:
The most comprehensive global directory of e-participation tools, projects and references. First launched in September of 2009, the site went dark around 2021 due to maintenance issues, taking a decade of the field’s institutional memory with it. Scripted the catalog out of Archive.org snapshots and reverse-engineered the app: 396 tools, 327 projects, 276 references.
The social graph of German Schlager credits: who wrote what, who performed it, and where the data came from. Built and maintained by two AI agents running locally on a cron schedule: Crawler, which decides what is worth reading, and Sherlock, which reads it and records every claim it finds — including the ones that contradict what is already there, because a conflict is data rather than an error. 10K songs, 3K people, 24K credits and counting, each sourced and confidence-scored.
A peacebuilding initiative envisioning a Middle East (re-)connected by high-speed rail. Advanced georeferencing pipeline, turning scanned images of 19th- and 20th-century survey maps into usable geodata, with measured accuracy. Every record carries its source. Full dataset is open source.
A vocabulary trainer for advanced Spanish learners working toward the DELE C1: the top 10,000 high-frequency words. A short stratified assessment across frequency bands estimates where your knowledge stops — the cliff most learners hit past B1 — and the app works the long tail beyond it, then exports to whatever spaced-repetition tool you already use. A separate Python pipeline builds and ranks the word list. I’m the target user, building it against my own exam prep.
Scaled, multi-lane history timelines built for visual comparison — seeing what overlapped what. One-click import of structured data from Wikidata, and the whole app runs bilingual in English and Spanish.
Event operating system for OFU 2026, a virtual gathering I instigated in 2013. Self-serve partner and sponsor onboarding, participant-driven unconference session planning, volunteer management, event report generation.
Re-imagining the online survey: questions roll out over time, participants watch results emerge, and shifts in opinion get tracked. Agile, where the traditional survey is waterfall. The spec was stress-tested before it was built — an adversarial review of the requirements document that surfaced 58 issues, 14 of them serious enough to block, while they were still cheap to fix.