Scientific peer review
Independent experts assess methods, data and conclusions before a journal article, systematic review or guideline is published.
Private medical evidence library
Reviewed evidence.Cited answers.
Aimedis Health Wiki turns peer-reviewed literature, clinical guidelines and approved reference material into a private evidence library. Nothing is published without a documented administrator review, and AVA, the Wiki's evidence assistant, answers with numbered citations to the exact source passages: from your reviewed library first, with PubMed and public health websites for what a general question needs beyond it.
Private workspace for approved organizations. No public registration.
- 45file formats, from PDF to video
- 182medical specialties
- 12specialty languages
- 5AVA analysis modes
- 100%publications with recorded review notes
Why peer-reviewed
An AI is only as good as the evidence it may read.
Medicine moves fast, and not everything that gets published, let alone everything online, is reliable. Peer review is science's quality filter: before a study or guideline appears, independent experts examine its methods, data and conclusions.
The Wiki is curated from peer-reviewed literature, clinical guidelines and approved reference works, and it adds a second filter of its own. Nothing becomes searchable until an administrator in your organization has reviewed it and documented the approval.
The result is a corpus you can defend. Every library citation points to material whose origin, version, licence and review are on record, and PubMed or web sources are cited with their links.
Wiki review, documented approval
An administrator compares the extracted text with the original and checks provenance, licence and currency. Review notes are mandatory, and every decision enters the audit trail.
Published evidence
Eligible for AVA and for connected Aimedis applications, with its full history attached.
- 01
Upload
The original is stored with authors, dates, licence and evidence level.
- 02
Quarantine & scan
File signature and size are checked, a SHA-256 fingerprint is recorded and a malware scan runs.
- 03
Isolated extraction
Text is extracted in a disposable process without database or AI credentials. OCR and transcription run on private processors.
- 04
Index
Passages are embedded for semantic search, keeping page, slide and timestamp locators.
- 05
Review
An administrator checks extraction and original, then approves or rejects; review notes are recorded with every decision.
- 06
Publish
The new version goes live and its predecessor retires in the same database transaction.
- 07
Cite
AVA and connected apps can retrieve it — until its review date expires or it is retired.
AVA vs. public AI
Precision isn't a bigger model. It's knowing where every answer comes from.
General-purpose chatbots are impressive writers, but they answer from a vast, unreviewed mix of internet text, and they write just as fluently when the evidence is thin. AVA works the other way round: evidence first, answer second.
Public AI chatbot
Trained on the open internet
- Draws on a vast, unreviewed mix of internet text. The origin, date and quality of each fact stay unknown.
- Writes just as fluently when the evidence is thin or missing.
- Can produce references that look plausible but don't exist.
- Knowledge is frozen at training time, so withdrawn guidance can linger.
- No reviewer, version or licence on record for what shaped the answer.
AVA in Aimedis Health Wiki
Grounded in your reviewed evidence
- For general questions, searches your reviewed library, PubMed and public health websites such as MedlinePlus at once, and answers from your library first.
- Says the evidence is insufficient instead of filling gaps from memory.
- Numbered citations open the original passage, its version and licence.
- Retired sources and sources past their review date leave retrieval immediately.
- Every published source carries its reviewer, version, fingerprint and licence.
- Treats source text as evidence, never as instructions.
AVA · Evidence assistantAsk
Summarize the reviewed evidence on first-line therapy for condition X.
- 1Clinical practice guideline v3 · p. 14 · CC BY 4.0
- 2Systematic review, 2025 v1 · p. 6 · licensed
- 3Society position statement v2 · §4.2
AVA · Evidence assistantAsk
A question that no reviewed or trusted source covers.
The available evidence is insufficient to answer this question. Add or retrieve relevant, reviewed sources.
How AVA builds an answer
- 1
Reviewed library first
Published sources with a current review date and matching access rights come first. For general questions, PubMed and public health websites are searched at the same time, and the answer relies on your library before them.
- 2
Semantic retrieval
voyage-3-large embeddings with 1,024 dimensions find the closest passages. Weak matches below the similarity floor are dropped.
- 3
Evidence packet
Retrieved passages are length-capped and handed to the model as untrusted evidence, never as instructions.
- 4
Server-selected model
AVA uses Anthropic Claude Sonnet 5, selected on the server. Requests cannot override the model or endpoint.
- 5
Citation check
Citation numbers must match the retrieved passages. Out-of-range or incomplete answers are rejected.
- 6
Answer or abstain
You get a cited answer, or a clear statement that the reviewed evidence doesn't answer the question.
Citations show where a statement comes from. They don't prove that every clinical claim follows from the passage, and similarity scores are not confidence probabilities. AVA supports evidence review. It doesn't replace clinical judgment.
Inside the Wiki
Everything a medical evidence library needs.
Organized by specialty, in 12 languages
182 medical specialties and fields of practice, searchable in every language below, with right-to-left layout for Arabic. Each specialty collects its reviewed reference sources.
- enCardiology
- deKardiologie
- frCardiologie
- esCardiología
- itCardiologia
- ptCardiologia
- nlCardiologie
- plKardiologia
- trKardiyoloji
- arطب القلب
- zh心脏病学
- ja循環器学
45 file formats
PDF, Office and OpenDocument, EPUB, HTML and Markdown, data files and subtitles. Images are read by OCR, and audio and video are transcribed on private processors.
- .txt
- .md
- .markdown
- .csv
- .tsv
- .json
- .jsonl
- .ndjson
- .xml
- .html
- .htm
- .rtf
- .srt
- .vtt
- .doc
- .xls
- .ppt
- .docx
- .xlsx
- .pptx
- .odt
- .ods
- .odp
- .epub
- .jpg
- .jpeg
- .png
- .webp
- .gif
- .tif
- .tiff
- .bmp
- .heic
- .heif
- .mp3
- .wav
- .flac
- .ogg
- .m4a
- .aac
- .mp4
- .mov
- .webm
- .mkv
AVA, five ways to work with evidence
- AskAnswer a focused evidence question.
- EvaluateInspect support, limitations and review gaps.
- ContextUnderstand setting, population and applicability.
- Cross-checkCompare claims across cited sources.
- ResearchBuild an internal evidence brief and find gaps.
Provenance you can inspect
Originals are preserved with SHA-256 fingerprints, authors, dates, licences and evidence levels. Citations point to the page, slide or timestamp, and every revision keeps its history.
- Source
- Clinical practice guideline
- Version
- 3 published
- SHA-256
- 3f9a0c7d…b84e21b
- Licence
- CC BY 4.0
- Review
- notes on record · due 2027-03
- Supersedes
- version 2 · retired
Collection statistics
Deterministic reports with no AI involved: evidence levels, formats, publication years, overdue reviews, provenance gaps and duplicate files.
PubMed built in
Search NCBI PubMed and import bibliographic records for review, without copying abstracts or full texts.
Medical terminology
Look up clinical codes through your licensed terminology service, with the code system and licence kept in view.
Interoperable by design
Read-only FHIR R4 resources for documents, provenance and libraries, HL7 v2 message validation, a TypeScript SDK and an OpenAPI description for developers.
Evidence for the Aimedis ecosystem
Aimedis platforms retrieve published, reviewed excerpts and podcasts with their own revocable reference keys. Every result carries source, version and licence. Private storage links and patient records are excluded from the shared service.
- Aimedis Care Pro
- REST API for your server
Security & privacy
Private by design. Auditable by default.
Medical knowledge work needs the same care as medical data. The Wiki enforces access in the database, not just in the interface, and keeps a verifiable record of every decision.
Invitation only
No public sign-up. Every account is an approved administrator of an approved organization.
Two-factor authentication
Authenticator-app codes. Once enrolled, the second factor is required for every page, API call and database read.
Isolation in the database
Row-level security separates organizations, corpora and patient subjects in the database itself, not only in the interface.
Scanned before it is read
Uploads are fingerprinted and malware-scanned, then parsed in an isolated process without database or AI credentials.
Hash-chained audit trail
Uploads, reviews, publications and credential changes are recorded in a per-organization hash chain.
Patient data stays off
A segregated patient workspace exists, but processing stays disabled until each processor has been approved.
These safeguards support GDPR- and HIPAA-aligned operation. They are not a certification and don't replace your organization's legal, contractual and clinical sign-off.
Questions
Good to know.
Can anyone sign up?
No. The Wiki is a private workspace for approved organizations and their administrators. Request information below, and the Aimedis team will get in touch.
Does AVA replace clinical judgment?
No. AVA helps you find, compare and cite reviewed evidence. It does not diagnose or prescribe, and clinical decisions remain with qualified professionals.
What happens when the evidence isn't there?
AVA says so. General questions are checked against your reviewed library, PubMed and public health websites; when the evidence covers only part of a question, the answer names the missing part, and if none of them supports an answer, you receive an insufficient-evidence response instead of a guess, which is a clear signal to add or review sources.
Why add peer-reviewed literature rather than any available text?
Because an AI answer can only be as reliable as the material behind it. Peer review checks methods, data and conclusions before publication, and the Wiki's own documented review checks each source's extraction, provenance, licence and currency before it becomes searchable.
Which AI models does AVA use?
AVA uses Anthropic Claude Sonnet 5 with server-controlled settings. General reference research searches Wiki evidence, PubMed and public health websites such as MedlinePlus together and answers from Wiki evidence first. Selected-source and patient analysis stay within their authorized evidence scope.
Can our own AI or other Aimedis apps use the evidence?
Each Aimedis platform or approved application gets its own revocable, audience-bound reference key (1–365 days or never expiring). Every valid key can read all published, reviewed and licensed reference excerpts and podcasts, each with source, version and licence. Patient records and private storage links are excluded.
Access
By invitation only.
There is no public registration. Tell us where to reach you, and the Aimedis team will send you information about the Wiki and how your organization can get access.
Your email address is used only to answer this request.
Already an administrator? Sign in