Juliana McMillan-Wilhoit
Back to Work
01 F&T Labs · Public Health Sauk County · 2023–2024

50 interviews to find the real problem: not missing tools, missing policy.

We went in to help Sauk County solve an AI problem. What we built for them became something bigger. But first: 50 interviews, a policy framework, and one very specific food code chatbot.

Public Health Sauk County, Wisconsin Baraboo, WI HIPAA AWS Bedrock RAG Policy Design
Impact
30m → 5m
Routine task time at Sauk County Health Dept
50+
Public health professionals interviewed before building
5+
LLMs tested before committing to AWS Bedrock
HIPAA
Compliant + Sunshine Law-ready from day one
Listen

Sauk County Health Department came to us with an AI question. Before touching anything, I ran 50 interviews across local health departments to understand the actual landscape. What I found wasn't a tools problem.

Staff were copying protected health information into free consumer chatbots. Directors trusted AI outputs for grant writing without understanding the limitations. Legal teams didn't know the exposure. The technology wasn't the problem. The policy vacuum was.

"The most valuable thing I built before writing a single line of code was a policy framework. Health departments couldn't safely use any AI tool without one."
Build

Once the policy framework was in place, we built the tool: a HIPAA-compliant, RAG-based food code assistant for Sauk County's environmental health inspectors. Ask a question in plain language, get a citation from the actual food code in seconds — not 30 minutes of PDF hunting.

I hand-coded the first prototype in HTML — not to ship it, but to get something real in front of users fast. Then iterated side-by-side with inspectors: tightening prompt phrasing, aligning tone with departmental style, and discovering that inspectors wanted to use it for drafting follow-up emails too.

AI policy framework — built before the product, now a standalone training offering
HIPAA-compliant RAG architecture on AWS Bedrock — data stays within the county's own AWS environment
Food code chat interface — natural language questions, citations in seconds, full audit trail
Sunshine Law compliance layer — every AI interaction logged, survives public records requests
Ask Why

I tested 5+ LLMs against real public health tasks — grant writing, policy summarization, data interpretation, response drafting — and scored outputs blind. AWS Bedrock won for one non-negotiable reason: HIPAA compliance. It keeps data within your AWS org and makes it possible to sign a BAA. The test results didn't matter if the data couldn't stay inside the customer's environment.

"What we built for Sauk County became something bigger. But that only happened because we solved one real problem first — for one real client — and got it right."
Sauk County Food Code Assistant — login screen

The Food Code Assistant, deployed for Sauk County Health Department

Sauk County Food Code Assistant — Ask the Food Code chat interface

The chat interface: inspectors ask questions, get answers grounded in the actual food code

Impact
"It's like having a really smart colleague who has read every policy we've ever written and never forgets anything."

— Sauk County Health Department

Routine tasks that took 30 minutes now take 5. 83% of inspectors reported reduced workload. 67% rated the tool as very or somewhat effective in early testing.

That's not "AI made things faster" — that's inspectors getting hours back per week to do the work that actually requires a human. And it only happened because we solved the policy problem first.

What we built for Sauk County became the foundation for PH360 — F&T Labs' platform for AI in local public health. One client, one real problem, solved right. Then productized.

Tools & Stack
AWS Bedrock HIPAA Compliance RAG / Knowledge Base Multi-LLM Testing SAFe Sunshine Law Python Embeddings
Next project
The military didn't need a better database. It needed data standards that worked globally.
ENSITE