devs·lopital
lopital

health guidance,
in your language.

Understand symptoms, lab results and medication, and know when to see a doctor. Built carefully, and never a replacement for a real clinician.

start a consultationhow it works

English, French, Hindi and Swahili.

how it works

a careful pipeline, not a chatty oracle.

  1. 1

    Intake and triage

    Your message is read in any of four languages and checked for emergency warning signs before anything else happens. Emergencies get the local emergency number first.

  2. 2

    Grounded answers

    Replies draw on published medical literature and your own record, and cite their sources. No invented studies.

  3. 3

    Handoff, not diagnosis

    lopital never claims to diagnose. It explains, escalates, and points you to a clinician when you need one.

safety

built around what medical AI gets wrong.

hallucinations

Up to 83% in adversarial conditions for raw LLMs. lopital uses retrieval-grounded prompts, refusal patterns for low confidence, and post-generation gates.

bias

Demographic bias documented in 91% of medical-LLM studies. We log every interaction and run a bilingual eval suite per release.

scans

General LLMs interpret X-rays at F1 7-18%. lopital refuses to read scans and routes them to a clinician, every time.

liability

Clear "information, not medical advice" framing. Audit-trail on every reply for regulatory review.

for clinicians

a partner, not a replacement.

If you run a clinic, hospital or NGO programme in an underserved area, we want to hear from you. Every patient who needs human care gets a clean handoff with their conversation history.

get in touch