What happens when everyone has the mic?
At most conferences, the people on stage have the microphone. Speakers, panelists and keynote presenters share prepared ideas with the audience. Attendees may participate in a Q&A or complete a survey afterward, but much of the knowledge held by the people gathered in the room leaves with them.
2026 Build-A-Thon
On August 3-5, 2026, the 7th World Conference convened at the University of Minnesota in Minneapolis, MN, bringing together academics, scholars, policymakers, activists, artists and others to discuss remedies to racial and ethnic economic inequality. Websmith Studio was invited to participate in planning the Build-A-Thon alongside members of the AI4Afrika coalition. AI4Afrika is a coalition of some of the world’s leading scholars and practitioners across disciplines including performance art, research and technology, and I am a member of that community. Websmith Studio, AI4Afrika and Ask an Elder sponsored the Build-A-Thon, an interactive session that ran in parallel with the conference programming.
The goal for the Build-A-Thon was to collectively create something that offered value to the organization and the community attending the conference. The project started with two questions:
- What happens when everyone has the mic?
- Could we capture the knowledge and wisdom of the people gathered at one conference, preserve the context around their contributions, and begin building a language model specifically for that community?
What and Why?
First, let’s level-set with a definition of “language model.” A language model is a tool trained on vast amounts of text that learns how humans communicate, allowing it to understand questions, generate written content, and help solve problems in natural, everyday language.
Next, let’s establish our intention for this community-centered language model. I’ve been exploring personal and smaller language models trained on individual or organizational knowledge and lived experience, so this convening presented a great opportunity to extend that work to a cultural or community-centered language model. The community model generated for 7WC is intended to:
- answer questions using knowledge contained within the archive;
- connect answers to participant contributions and sources;
- preserve differing perspectives rather than invent consensus;
- recognize relationships across stories and ideas;
- retain relevant participant context; and
- understand the boundaries of what it knows.
Establishing Ground Rules
We also established a few ground rules that reflect our guiding philosophy for how these projects should work and how contributions would be handled.
- Participants were told what we were building and how their information would be used before providing consent.
- Participants were asked to indicate their preferred name and enable the appropriate spelling of their name.
- They could choose how they wanted to contribute.
- They could see where their contribution would live.
- They can provide updates, new information or ask for information to be removed at any time.
- They can see any model-generated insights gleaned using their data.
Getting Started
We arrived in Minneapolis with a blank project web page and an existing set of tools Websmith Studio had already built to automate the pipeline that ingests and processes the data needed to train the model. So that we could focus on the most important part of the process, sharing stories, the automation pipeline was key. It enabled us to focus on having and capturing conversations with participants, while the routine automatically processed, organized and surfaced the stories as they were recorded.
Natural language processing scripts parsed the text to pull out details that are important for contextualizing the audience, such as the birthplaces and locations of the members of this diverse global community. Transcripts were automatically categorized under the core thematic pillars of the conference whenever those topics were raised. Our indexing routines flagged recurring patterns, shared perspectives and cross-cutting observations across stories without forcing rigid topic boundaries.
This setup turned raw media into a dynamic, searchable archive that connects related ideas across different interviews and highlights shared insights. The Build-A-Thon gave us an opportunity to apply this framework in a real-world setting, and the resulting Living Archive is now hosted on Websmith Studio infrastructure and available to community organizers and participants.
Collecting Stories
We also established a few ground rules that reflect our guiding philosophy for how these projects should work and how contributions would be handled.
- Participants were told what we were building and how their information would be used before providing consent.
- Participants were asked to indicate their preferred name and enable the appropriate spelling of their name.
- They could choose how they wanted to contribute.
- They could see where their contribution would live.
- They can provide updates, new information or ask for information to be removed at any time.
- They can see any model-generated insights gleaned using their data.
The Outcome
The result was “A Living Archive of Community Wisdom“, a working archive of this community’s stories that can be searched by context, connected across related ideas, analyzed for emerging patterns and explored geographically.
Over the three days, we delivered a working version of the Living Archive with the following core elements in place:
- Model scope and behavior defined – Including attribution, differing perspectives and knowledge boundaries.
- Stories collected and preserved – Across video, audio and transcripts.
- Participant context captured – Including key personal, geographic and conference-related information.
- Archive indexed – Contributions connected and searchable by meaning.
- Context grounding rules established – Insights tied directly to relevant source material.
- Insights surfaced – Recurring ideas and relationships can be examined across the collection and traced back to supporting stories.
Training & Fine-tuning
The next phase of this project involves training and fine-tuning this small, open-weight language model using the community data archive. We are evaluating it against questions it has not previously seen, and the model will continue to evolve as more contributions and feedback become available.
As this work progresses, we will launch an update on our website allowing 7WC participants to prompt the model directly and review its outputs. The archive remains open to the 7WC community for continued contributions, feedback, and updates.
If your community or organization is interested in learning more about this project, contributing to this work, or exploring ways to collaborate, please feel free to reach out.
Connect
Websmith Studio works at the intersection of craft and imaginative technologies, creating original work across mediums, from software and interactive systems to visual art, structured play, and narrative form.
Interested in one of our projects, assets, or ideas? We welcome inquiries from people and organizations who want to learn more, explore a potential collaboration, or contribute to work already underway.