AI for Functional Medicine

Why does generic AI fail at functional medicine documentation?

Because it was built for something entirely different. Every major AI scribe on the market — Nuance DAX, Suki, Abridge — was designed around the 15-minute sick care visit. The bread-and-butter of their training is the problem-oriented medical note: chief complaint, assessment, plan. That's what their models do well.

Functional medicine documentation doesn't work that way.

A 90-minute initial consult generates dozens of data points across multiple body systems, dietary patterns, environmental exposures, and lab interpretations. You're tracking DUTCH hormone panels, OAT results, GI-MAP stool analyses, and BaleDoneen cardiovascular risk assessments. The "assessment" isn't a single diagnosis — it's a systems-biology synthesis that connects the dots between elevated urinary cortisol metabolites, zinc deficiency, and mitochondrial dysfunction.

Generic AI can't handle this because it was never asked to. It doesn't understand what a DUTCH test tells you about HPA axis regulation versus a single serum cortisol. It doesn't know that an OAT result showing elevated quinolinic acid with low serotonin metabolites points toward a completely different treatment pathway than elevated succinic acid alone. These aren't edge cases in functional medicine — they're the baseline.

What makes functional medicine documentation uniquely hard?

Three things that no general-purpose AI can handle:

First, the documentation requires longitudinal pattern recognition. You're not just documenting today's visit — you're showing how this week's lab results connect to symptoms reported three months ago and supplement changes made six months before that. Generic AI sees each note as a standalone document. HANS sees your patient's entire clinical narrative.

Second, the language is different. When you say a patient has "MTHFR with COMT +/+ and slow methylation," that's clinically meaningful. When you say they have "genetic SNPs affecting folate metabolism and neurotransmitter degradation," that's accurate but verbose. Generic AI tends to flatten functional medicine terminology into generic phrasing that loses clinical precision — or worse, invents information to fill gaps it doesn't understand.

Third, the scope of review is massive. In a single functional medicine visit, you might review 15+ labs, discuss nutritional biochemistry, adjust 5-7 supplements, and document lifestyle factors that would never fit in a conventional SOAP note. The documentation volume is 3-5x what you'd produce in a sick care visit, but the time you have is the same.

Can generic AI at least handle basic transcription accurately?

Basic transcription? Sure. But accuracy in recording words isn't the problem — it's relevance and clinical context that matters.

Here's what happens in practice: you dictate a 20-minute case review covering a patient's DUTCH results, recent diet changes, and new symptoms. Generic AI transcribes every word accurately. Then it produces a note that says "discussed DUTCH test results, reviewed diet, patient reports improved sleep."

That's not useful. A functional medicine note needs to capture that the patient's elevated evening cortisol on DUTCH correlates with her report of midnight anxiety, that her recent elimination diet resolved SIBO symptoms but may have created protein insufficiency, and that the new supplement protocol addresses both the cortisol dysregulation and the protein concern. That's what makes the note clinically actionable at the next visit.

Generic AI gives you transcription. HANS gives you documentation that actually helps you practice.

Does it matter if the AI understands the difference between different functional medicine lab panels?

Absolutely. Here's a concrete example:

On a DUTCH panel, an elevated 11ß-HSD activity ratio suggests excessive cortisol breakdown, often seen in patients with high stress loads and blood sugar dysregulation. On an OAT, elevated 11ß-HSD would be a completely different finding — the OAT doesn't even measure cortisol metabolites in the same way.

Generic AI can't distinguish between these because it was never trained on what these tests actually measure. It sees "11ß-HSD" and either ignores it or produces a generic description that loses all clinical meaning.

HANS knows the difference. We've trained our models on real functional medicine labs so they understand that when you're reviewing a DUTCH panel, certain patterns indicate HPA axis dysregulation requiring specific interventions, while the same findings on an OAT would suggest an entirely different clinical pathway.

What does HANS do differently?

HANS was built from the ground up for functional medicine. We trained our models on actual functional medicine charts — not primary care notes adapted for FM, but the real thing. DUTCH interpretations. Organic acids analysis. The Bredesen protocol. IFM-style case presentations.

When HANS processes your patient prep, it knows that an elevated 5-HIAA on a 5-HIAA/creatinine ratio isn't just "elevated serotonin metabolism" — it's a marker that might indicate mast cell activation, small intestinal bacterial overgrowth, or medication interaction, depending on the clinical context. HANS surfaces these connections because we've taught it to think like a functional medicine practitioner, not a transcriptionist.

The result: 2 hours of patient prep compressed into 10 minutes. Notes that are clinically precise, not generically sanitized. And notes that actually help you practice better medicine because they surface patterns you might have missed.