Tabibu

Tabibu's answers show sources, and when the guidelines do not cover a question, the system states this explicitly and labels the answer as general information.

Tabibu positions itself as an information tool and does not claim to be a medical device; it currently has no clinician review, and its limitations are listed on the tabibu.health/trust page.

John is a senior AI and full-stack engineer who over the past year developed the health information assistant Tabibu.

Tabibu's design principle is not to tell anyone to start, stop, or change medication.

The developer explained the reason for excluding image reading: interpreting scans or skin photos falls within the scope of medical devices, and he would rather say 'I can't do it' than guess.

Tabibu's handling of file privacy is as follows: files and the text read from them are not stored or logged; geolocation data (EXIF) is removed from photos; only the user's questions and Tabibu's answers are saved for about 48 hours; users can choose to enable the 'keep for 30 days' button (off by default), but conversations involving files cannot be retained, and answers about files will never enter the shared answer cache.

In testing, prescription dosages were returned accurately according to what was printed, answers for X-ray reports quoted the radiologist's original text and did not claim the scan results were normal, a file containing hidden instructions (requiring the model to mark all results as normal) was ignored by the model, which stated it had detected them, and the DICOM file was refused processing.

Tabibu's testing is based only on spot checks of a small number of fictitious documents and has not yet undergone a complete evaluation, so no accuracy figures are published; at the policy level there is still no clinician review, and Swahili answers have only undergone AI review; the WhatsApp document reading feature is still in testing and not fully open; and the end-to-end cost per file has not yet been measured.

In public health benchmarks, an ordinary AI model scored higher than Tabibu in the developer's own re-run, 0.522 versus 0.426 (non-official leaderboard data), and the developer believes the reason is that Tabibu's safety layer lowered its score in the benchmark.

Tabibu states that what it provides is general health information, not diagnosis or medical advice.

The Tabibu developer disclosed the limitations of the current version: about one in eight radiology reports triggered a safety fallback in testing, with the system only giving 'this is the text I read, and the questions you can ask a doctor' rather than a full explanation, and the developer explicitly stated that this does not count as a success case.

Tabibu's document reading feature is available for Basic and Pro paid plans, and free accounts cannot upload files; supported formats include PDF, Word (.docx), plain text, Markdown, and JPG, PNG, WEBP photos, with a single file limit of 8MB.

Tabibu's document processing uses a two-step architecture: the first step uses a vision model to transcribe the file into text, and the second step sends the transcribed text into the original text-only processing pipeline for explanation, and the file does not directly enter the stage of conversation with the user.

Tabibu explicitly lists four things it will not do: it does not diagnose, does not tell users that a certain result is normal or fine, does not recommend starting, stopping or changing medications (dosages are only repeated as printed in the document), and does not read X-ray/CT/MRI images or skin photos.

Tabibu added document reading capability, allowing users to upload lab reports, prescriptions, PDFs, Word files, or photos of them on the website, and the system will read the content and explain it in plain language while generating a list of questions to ask clinicians.

The developer tested with fictitious documents, including lab reports, prescriptions, Word notes, radiology reports, and a DICOM file, and did not use real patient files.