aviv hidrian. studio of one.
amsterdam, since 2019. physiotherapist by day, AI web developer by night. one person, multiple disciplines, no handoffs.
how i got here.
i didn't go from physio to development. it was the other way around.
i started as a web developer and database programmer. code was rewarding but i missed working with people. so i trained as a physiotherapist and started practicing in 2008. hands-on work, real human feedback, immediate results.
after a few years in physio, i missed the building. i also noticed how badly most healthcare practices were served by the technology around them. software designed without seeing the workflow. websites built by agencies who'd never set foot in a treatment room.
so i came back to development, with healthcare knowledge baked in. in 2019 i started avivly as a studio for that intersection. design, development, technical seo, healthcare ai, all in one ecosystem so i never have to compromise on quality.
i still practice as a physiotherapist. i also teach entrepreneurship at the european school of physiotherapy in amsterdam, where i help students think about how to actually run a practice, not just treat patients.
what that means in practice: when i build a website for a fysio practice, i understand both the website and the practice. when i build an ai tool for a clinic, i know what the workflow actually looks like. when i teach, i bring real cases from both sides.
one person. multiple disciplines. no handoffs.
a clinical operating system for chronic low back pain.
i'm aviv — physiotherapist and builder. pt-charlie started with one question: can we guide people with chronic low back pain earlier, more safely, and more humanely — without replacing the physio?
in 2024, this prototype won the elaa verBETER-prijs, the award for the best innovative healthcare project in amsterdam & almere.
the elaa verBETER-prijs recognises primary-care innovations that make a measurable difference in the first line of dutch healthcare. pt-charlie won for bringing together clinical decision trees, language models, and empathic design — built on a fine-tuning corpus of >288,000 tokens spanning 400+ clinical scenarios.
→ watch the winning 4-minute pitchwhat pt-charlie actually is.
most people describe it as "a chatbot with 47 indicators." that's only the foundation. seven layers sit on top of it:
47 peer-reviewed multidisciplinary indicators — somatic, psychosocial, pain behaviour.
21 self-developed clinical profiles that translate raw indicators into a working picture.
4 routing dimensions: pain characteristics, structural changes, psychological factors, functional limitations.
7-question safety screen — cauda equina, fracture, malignancy, neurological deficit.
3 deterministic tiers — detects crisis language, overrides the LLM on suicidal signals, surfaces 113 Suicide Prevention immediately.
Daily check-ins on four WHO ICF-anchored activities with trend lines for the clinician.
7 empathy principles, a banned-phrases list, and a "two-bubble rule" — warmth as a system, not a prompt trick.
together: 80+ deterministic clinical artifacts, plus a workplace ergonomics photo-scan, a voice mode, case-based reasoning over a semantic patient-case library, and a transparency architecture that strictly separates text generation from clinical logic — the LLM handles only the conversation, never the clinical judgement.
scientific foundation.
the 47-indicator framework is adapted from "the nijmegen decision tool for chronic low back pain" (van hooff et al., plos one, 2014), published under creative commons cc by 4.0 — an open-access licence that explicitly permits reuse, adaptation, and commercial application with attribution. everything built on top — the archetypes, the safety layers, the tone engine, the architecture — is original work.
what it explicitly is not.
- not a medical device: no ce mark, no mdr classification.
- not a replacement: for a physiotherapist or physician.
- not an autonomous diagnostician: every decision is readable, traceable, and human-overridable.
why i build this.
because half the people with chronic back pain get stuck before they find the right care.
because a physio never sits next to someone at 11 PM who can't sleep from the pain.
and because AI shouldn't imitate the clinician — it should make the waiting-room moment, the in-between moment, the silent moment, better.
that's what pt-charlie tries to do: a calm, clinically grounded voice during the hours no one else is around.
pt-charlie is an experimental developer prototype and concept study, not a clinical medical device. its screening structure is adapted with attribution from the 47 multidisciplinary indicators established in "the nijmegen decision tool for chronic low back pain" (van hooff et al., plos one, 2014, cc by 4.0). its clinical rules serve only to demonstrate how language models can navigate evidence-based decision trees, and must never substitute for diagnosis or judgement by a licensed clinician.
what i bring.
- design. every site i ship, i design.
- develop. i write the code myself. clean wordpress, no plugin pile-up.
- ai for healthcare. applied ai, real workflows. pt-charlie is the working example.
- teaching. entrepreneurship at the european school of physiotherapy.
- practice. physiotherapist since 2008. i live in the world i build for.
the way i work.
first meeting is always free. 30 to 45 minutes. i look at your site, your goals, your competitors. you leave with a clear picture even if that picture is "you don't need me right now."