Medical professionals - clinical knowledge

A physiotherapist sees 15 patients a day. Over years, they build up experience — which stretches help with which conditions, which exercises work for post-surgical recovery, which nutritional advice they keep giving to the same type of patient. That knowledge lives in their head, in scattered notes, and hardly anywhere else.

A chiropractor has the same problem. So does a nutritionist. So does a coach preparing an Olympic athlete.

They all share two characteristics:

1. Their accumulated knowledge is valuable. The treatment protocols, the exercise progressions, the nutrition plans that actually work — this is expertise built over years of practice.
2. Their data is sensitive. Patient health records are legally protected. An athlete's training program is a competitive secret. In both cases, the information cannot leave the premises.

lokaliQ turns that scattered expertise into a searchable knowledge base — one that runs on hardware you control, with encrypted storage, and never sends data to a cloud service.

What Goes In

The practitioner uploads their own material. Not patient records directly — treatment protocols, clinical guidelines, exercise libraries, nutrition frameworks, research papers, and anonymized case notes.

For a physiotherapist or chiropractor:

For a sports performance coach:

The domain structure depends on how the practitioner thinks about their work. The point is separation — rehabilitation research shouldn't drown out practical exercise descriptions in the search results.

Domains marked for Quick Notes also accept short text entries directly from the sidebar — ideal for the case_notes domain, where observations are captured in the moment rather than written up as formal documents.

How It's Used

The practitioner is with a patient. They need a treatment plan for a 45-year-old with chronic lower back pain, mild disc herniation, sedentary job.

They query the system:

"Treatment protocol for chronic lower back pain with L4-L5 disc herniation, sedentary patient, no surgical history."

lokaliQ retrieves across domains — the relevant protocol, the exercise progressions that apply, any research on conservative management of disc herniation, nutritional recommendations for inflammation, and similar past cases where this approach worked.

The practitioner gets a structured response they can adapt to the specific patient. They're not starting from scratch. They're starting from their own accumulated practice.

A coach preparing a training block for a middle-distance runner asks:

"8-week base building phase for 1500m runner, coming back from a stress fracture, current VO2max estimated at 62."

Same principle. The system pulls from the training domain, the rehab domain, and past athlete histories with similar profiles. The output is a starting point, not a finished plan.

Why This Data Cannot Leave the Premises

Health data is governed by strict regulations — GDPR in Europe, HIPAA in the US, similar frameworks elsewhere. Patient information must be stored securely, with controlled access, and cannot be transmitted to third-party services without explicit consent.

Most cloud-based AI tools are immediately disqualified. The moment you type a patient's symptoms into ChatGPT, that data has left your control.

lokaliQ runs on your own hardware. The LLM runs locally via Ollama. The documents are stored in an encrypted volume that requires authentication to mount. The embeddings and vector indices stay on the same machine. Nothing leaves the building.

For sports performance, the motivation is different but the requirement is the same. A national athletics federation's training methodology is a competitive advantage built over decades. A coach's periodization strategy for their Olympic sprinter is not something they want indexed by a cloud provider. Self-hosted, encrypted, air-gapped from the internet — that's not paranoia, it's professional responsibility.

The Pattern Reuse Problem

Here's what makes this more than just document search.

A physiotherapist treating their 50th shoulder impingement case doesn't need to look up what shoulder impingement is. They need to know: What have I done before that worked well for this type of patient?

The case_notes domain is where this happens. After each treatment cycle, the practitioner writes a brief anonymized summary: patient profile, presenting complaint, treatment approach, outcome. Over time, this builds into a body of evidence — not clinical trial evidence, but practical evidence from their own practice.

The barrier to building this evidence base has to be low. If capturing an insight requires creating a file, uploading it, and running an ingestion pipeline, it won't happen between back-to-back appointments. lokaliQ's Quick Notes feature removes that friction. The practitioner opens a sidebar panel, selects the target domain (e.g. case_notes), types a few sentences, and hits save. The text goes straight through chunking and embedding into the knowledge base — no file, no extra steps. Two minutes between patients is enough to capture what worked, what didn't, and what to try differently next time.

When they query the system for a new patient with similar characteristics, the case notes surface alongside the formal protocols and research. "Three similar patients responded well to this exercise progression combined with this manual therapy approach" is more useful than a generic protocol from a textbook.

This is the practitioner's expertise, externalized and made searchable.

Structured Output for Treatment Plans

The same structured template approach from the public building permit case study applies here. Instead of free-form text, the large LLM populates a template:

{
  "assessment": "Brief clinical reasoning",
  "recommended_protocol": "Protocol name or description",
  "exercises": [
    {
      "name": "...",
      "sets_reps": "...",
      "progression": "...",
      "contraindications": "..."
    }
  ],
  "nutrition_notes": "...",
  "similar_cases": ["Summary of relevant past cases"],
  "review_in": "4 weeks",
  "red_flags": ["Symptoms that warrant referral or imaging"]
}

The practitioner doesn't get an essay. They get a structured plan they can review, adjust, and apply. The template ensures consistency — every treatment plan covers the same bases, every time.

What This Doesn't Do

This is a knowledge retrieval tool, not a medical device. It does not:

• Diagnose conditions
• Replace clinical judgment
• Generate treatment plans autonomously
• Constitute medical advice

The practitioner queries their own knowledge base, gets back suggestions based on their own documents and past practice, and makes all clinical decisions themselves. The system surfaces information. The practitioner decides what to do with it.

Security Summary