On October 8 we hosted AI for Environmental Health with PH360™, a webinar for the people who handle inspections and permits. A guest from a rural Missouri health department showed how he uses PH360 every week. Then Jefferson and Brett ran a live demo of Spruce, our environmental health permitting and inspection app, from the first permit application to the issued permit. If you missed it, the 32-minute recording is below, with a recap of what we covered.

General-purpose AI guesses. Your inspectors can’t.

Cailin Walker, a business analyst on our team and a former health inspector, opened with a situation most inspectors know. You’re in the field, you hit something you’re not sure about, and you don’t want to dig through your whole code book. So you ask ChatGPT. In her example, a question about freeze-dried strawberries and eggs under the FDA Food Code, ChatGPT answered from Illinois rules instead. PH360 answered from the 2022 Food Code and showed where the answer came from.

That’s the problem in environmental health: every jurisdiction has its own code, its own policies, and its own way of doing things. AI only helps on the job when it knows your rules. PH360 puts that AI in three places:

• Maple knows your code and answers questions from it.
• Oak learns your rules and adds them to what Maple knows.
• Spruce runs the field work: permits, inspections, citations, and reports.

How one rural health department uses it every week

Jesse DeVillier wears a lot of hats at a small health department in Missouri. He built an Oak agent on his state’s Environmental Health Operation Guidelines, along with the linked attachments like home-based kitchen guidance, so it answers from Missouri’s rules, not a generic code.

He tested it on a real case from his farmers market: a vendor selling fresh eggs, freeze-dried eggs, freeze-dried strawberries, and freeze-dried candy. Freeze-dried food is a gray area. In Missouri, the guidance lives in an email that leaves the call to the local inspector. The agent came back with a table that sorted each product, cited where each answer came from, and said plainly that the final call was his. Then he checked it himself, and it matched.

He now uses it for a weekly scenario quiz with his inspectors, so nothing surprises them in the field. He’s also built scenarios from real calls, like a wrecked delivery truck carrying food earlier that week. He uploaded his county’s top violations and his food handler course, and it wrote ten new test questions aimed at those violations, with a table mapping each question to the violation it covers.

Spruce, start to finish

Jefferson McMillan-Wilhoit, our CEO, explained the order we built things in. Maple and Oak came first because if AI couldn’t be safe and reliable in public health, nothing else mattered. With that foundation in place, Spruce uses AI where it actually helps inside the environmental health workflow. Then Jefferson and Brett Emo, a principal consultant at F&T Labs who used to teach environmental health, walked one restaurant from application to permit.

1. The business applies online. Elena, the owner of a new restaurant, signs in to the community portal. Spruce already knows her and her establishments, so she isn’t re-entering the same information. The questions adapt as she answers. Switch the category to retail food and the seating count question disappears. She types her rough hours in plain language, uploads the required documents, and pays the fee. Her progress saves as she goes, so a missing menu doesn’t mean starting over.
2. Spruce prepares the review file. When the application comes in, Spruce checks that the documents roughly fit the requirements and suggests a risk classification. That suggestion shows up in a purple box. In Spruce, purple means AI drafted it and a person decides.
3. The application goes to review. Dana, the inspector, assigns it to herself, and the status updates without extra clicks. She can send a document back for resubmission, then request the pre-operational inspection.
4. The inspector plans the day. Routine inspections, follow-ups, and pre-op requests all show up in one list. Spruce suggests an order for the day based on risk category and distance from the starting point. The inspector accepts it or changes it.
5. The inspection happens on an iPad or iPhone. At the restaurant, the field app opens with a briefing: what was cited last time and whether those items were resolved. The marking sheet follows the FDA format (in, out, not observed, not applicable, corrected on site), with your local code added and given priority.
6. The inspector writes the observation, and Spruce drafts the citation. Sliced tomatoes at 47°F in the walk-in become a drafted citation with the code section, the marking sheet item, a plain-language explanation, a corrective action, and a correction deadline. Priority levels don’t come from AI. They’re tied to the code section, so a priority item is always a priority item. When an observation could fit several sections, like scratched and stained cutting boards, Spruce lists the options and the inspector picks one.
7. Inspection to report to permit. A pre-finalization checklist, a plain-language summary for the person in charge, and signatures from both sides close out the inspection. Spruce creates the follow-up visit based on how serious the violations were. All it needs is a day and time. The report is built in your department’s format. In the demo we matched a Massachusetts town’s public report, with temperatures included. When the pre-op passes, Dana approves the application, the permit is issued with its conditions, and Elena sees it in her portal.

What it won’t do

Jefferson typed in a question instead of an observation, asking whether he could keep the mop sink inside the walk-in. Spruce declined to write a citation. A question with no observed facts isn’t something to cite, and Spruce is set up to wait for enough context before it suggests one.

Brett pointed to one change it could drive. When inspectors focus on writing good observations and the code citations get picked up for them, new inspectors don’t need an encyclopedic knowledge of the code to feel confident in the field. That’s a real difference for retention, and for how we train the next generation of inspectors.

---

Questions about Spruce

Can we load our existing inspection data into PH360?

Not automatically yet. Spruce is built first for departments moving off paper. Importing records from an existing inspection system is on the roadmap, and we expect to have it relatively soon. Until then, bringing in past records means entering them by hand.

Can businesses request permits through PH360 themselves?

Yes. Your department sets up the application, including the questions, the required documents, and the fees, and community members register in the portal to submit applications, upload documents, and pay.

What happens when we adopt a new edition of the FDA Food Code?

You set a launch date for the new edition in your configuration. Past inspections keep pointing to the edition they were cited under, so no old citation gets silently renumbered, and new inspections use the new edition from that date on. PH360 comes with recent FDA Food Code editions preloaded. If your department uses an edition we haven’t preloaded, it’s handled the same way as a local code modification, and you configure it without needing us.

Does the AI decide whether something is a violation?

No. Spruce drafts the citation, the plain-language explanation, and the corrective action, and the inspector reviews it and decides. Priority, priority foundation, and core levels are fixed to the code section, not judged by AI.

Thank you to Jesse DeVillier for sharing his work, and to Cailin Walker and Katelyn Wolf, both former health inspectors, for hosting with us. For more on asking questions of your food code, read AI for Environmental Health: Chat With Your Food Code, or see everything in PH360.