AI Has a Rural Blind Spot. This Summer I Helped Build a Fix.

This summer I was one of twelve Colorado Thrives AI Fellows, a new program run with Anthropic and CodePath that pairs students with Colorado organizations for a summer of real AI work. I did mine through my employer, the Center for Rural AI. My colleague Emaliah Sawyer and I were CRAI's two fellows. The fellowship gave me weekly mentorship through CodePath, hot-seat sessions in front of the other fellows, technical reviews with Anthropic staff, and a showcase in Denver at the end. Most of all it gave me a project to carry from idea to something the team uses every day.

The Idea I Brought In Was About Accessing Data Related to Rural Communities in Regards to AI.

AI models learn from what has been written down and published online, and rural America is thin in that record. Over 50 million people live in areas that amount to a training-data blind spot. Ask a model about a rural county and you get a shallower, less accurate answer than you would for a city. The cost lands on real communities, from urban-trained medical AI tools misfiring in rural clinics to small farms left out of tools built for large operations. I wanted to build a way for rural organizations to get at the information the models were missing.

Then Anthropic technical staff took that idea apart. They helped me realize that data access actually wasn't the primary problem. Instead it was that the knowledge CRAI had already collected had nowhere to live. CRAI's team had been gathering rural AI information for months: articles, grants, legislation, research, stories. But it was scattered across files, inboxes, and individual people's laptops. Every time someone needed it, they went and gathered it again. When someone was out, their knowledge was out with them. Before CRAI could offer anything to other organizations, it needed an internal tool that solved its own scatter. The knowledge existed. There was nowhere centralized to put it that the AI could reach.

That reframing changed the project, and it changed how I build. I learned to target the root of the problem instead of defending a solution I walked in with.

What I Built

BRAIN stands for Bedrock for Rural AI kNowledge. It is one curated library of rural-AI information, all vetted and stored in one place. Every answer it gives carries a citation back to the source. And it plugs directly into the AI tools our team already uses, so when someone at CRAI asks Claude a question, Claude can pull from BRAIN instead of guessing from general training data.

That is the Whole Idea. Give the AI Rural Ground Truth, Built In.

Today BRAIN holds more than 100 distinct sources, ingests new material automatically around the clock, and the team uses it every day. The piece I'm proud of is that it works as a plug-in rather than a separate app. Nobody has to open a new tool or learn a new workflow. They just ask in the session they already have open, and the answer comes back sourced.

Here is what that looks like in practice. I asked BRAIN about water usage in rural communities and, in about thirty seconds, got back current, cited data and specific stories from actual places. Ask a plain AI model the same question and you get something general and vague, because the model has never really seen those places. BRAIN has.

Emaliah tested this. She built BRAIN Bench, a hundred-question evaluation that scored six AI setups on accuracy, completeness, rural nuance, and honesty. BRAIN came out ahead on accuracy and honesty and behind on completeness, because it gives short, sourced answers instead of long ones. The lead is small enough that we're calling it a first measurement, not a win. But it confirmed the thing I care about most: for a grant writer or a clinic director, a confident wrong answer costs more than a short one. Citations are paramount in trusting these models and using them effectively.

What The Fellowship Taught Me

I gained immense technical proficiency from the training with Anthropic technical staff. However, my biggest takeaway was about communication: sharing ideas clearly, highlighting problems, and explaining my struggles. In the brainstorm phase I learned to stop designing alone and ask people what CRAI's team actually needed. Weekly mentorship with Cam at CodePath taught me to say what I was actually stuck on instead of talking around it. Hot-seat sessions taught me to defend a design to people who would push back. The Denver showcase, five minutes, taught me to explain the work to non-technical people who don't want to focus on how it's built, rather why it matters.

Because of this fellowship, I've realized that organizations like CRAI, CO Thrives, CodePath, and Anthropic can benefit if they work together to help elevate each other's missions. I luckily got to be the one elevated through this collaborative program.

Where It Goes

BRAIN was proof that a small nonprofit could build a curated, cited knowledge layer and put it where the AI can reach it. The work happening at CRAI now takes that same idea down to the county level.

The team is building a system called Lodestone that pulls public data, source by source, for a specific rural county: economics, agriculture, health, workforce, and more. Twenty-two sources for a single county now run in about twenty minutes. As of this week it covers a dozen counties and more than a million data points, starting with places in North Carolina and the Midwest, and it is growing.

Going county by county changes how the blind spot looks. It stops being a national statistic and becomes a specific, fixable gap. In federal farm data, the counties around Durango have hundreds of entries each. San Juan County, next door, has one. Filling in that missing layer is most of the work ahead, and it is the kind of work BRAIN showed a small team could do.

The takeaway for me is simple. The models aren't necessarily the hardest part right now. The hard part is collecting what rural America actually knows, one county at a time, with sources, and putting it where the AI can reach it.

Holden Bronson is a Product Builder at the Center for Rural AI, was a 2026 Colorado Thrives AI Fellow, and is a sophomore studying Computer Science at Columbia University.