The evidence engine Cynaps Chat · Preview
Answers you can verify, not just read.
Ask in clinical language. Get a synthesized answer where every claim resolves to the paper, guideline or chapter it came from — drawn from 36 million peer-reviewed articles, ICMR and AIIMS guidance, society statements, trial registries and standard texts, re-indexed every day.
Cynaps Chat · Preview
Reply...
1 / 9 · Every claim carries its source
Decision support · not a diagnosis
Fluent is easy. Verifiable is the hard part.
Checkable
Every claim carries its source
Answers are written from retrieved passages, and each assertion resolves to the paper, guideline or chapter it came from — with the reference list kept separate from the generated text so you can audit provenance at a glance.
Current
Re-indexed every day
New literature is pulled and indexed daily, so an answer reflects what was published this week — not a model's training cut-off. On fast-moving guidance that difference decides whether an answer is right.
Honest
It says when the evidence is thin
When the corpus doesn't support a confident answer, the pipeline is built to say so rather than fill the gap with fluent prose. Refusing to guess is a feature, and it is the behaviour a general-purpose chatbot is worst at.
Inspectable
The reasoning is visible
You can open how the answer was reached — what was searched, which passages were used, where the numbers came from — instead of taking a paragraph on trust.
How an answer gets built.
Retrieval is the product. The model is the last step, and it only gets to write from what the search found.
Search, two ways at once
Your question runs through a medical-domain embedding model that matches on meaning, and a learned keyword model that catches the exact drug, gene or scale name. The two result sets are fused, so neither paraphrase nor precise terminology gets lost.
Rerank for what actually answers it
A reranker re-scores the candidates against your specific question. This is the step that pushes the merely topical out of the way of the genuinely relevant.
Answer only from what was retrieved
Generation is constrained to the retrieved passages and instructed to attach a marker to each claim, so the reference list is a record of what was actually used rather than a bibliography added afterwards.
Check the numbers
Doses and thresholds are checked back against their cited source; a figure that isn't supported by the passage it sits next to gets flagged rather than passed along quietly.
What it reads, and how often.
Peer-reviewed literature
36 million articles, with open-access full text indexed at multiple granularities so a dosing detail and a synthesis question can each be answered at the right depth.
Indian guidance
ICMR and AIIMS guidelines, indexed alongside the international literature so answers can reflect practice here rather than defaulting to Western protocols.
Society statements
Guidance from bodies including WHO, NICE, IDSA, IAP, ISN, ADA and RSSDI.
Trial registries
Registered clinical trials, so questions about ongoing or recently completed studies are answered from the registry rather than from memory.
Standard textbooks
Reference texts for the settled fundamentals a journal article assumes you already know.
What it is — and what it isn't.
Cynaps is clinical decision support for licensed professionals. It retrieves evidence and drafts synthesis; it does not examine your patient, and it does not replace your judgement. Verify high-stakes recommendations against the primary sources and your local protocols — the citations are there so that check takes seconds rather than an afternoon.
Common questions
How is this different from asking ChatGPT or Gemini?
A general-purpose model answers from its weights and, at best, a live web search. Cynaps answers from a medical corpus it retrieved for your specific question, and shows you which passages it used. The practical difference shows up at the edges: on questions where the literature is thin or very recent, an ungrounded model tends to produce fluent, confident, unsourced text.
Where do the citations come from?
From the passages actually retrieved for your question. The reference list is assembled from what the answer was written from, not generated alongside it — which is why the markers resolve to real records you can open.
How current is it?
New literature is indexed daily. Guidelines and textbooks are re-ingested when their sources publish updates.
Can I use an answer as the basis for a clinical decision?
No. This is decision support for a licensed clinician, not a diagnosis and not a substitute for your own judgement. Verify high-stakes recommendations against the primary sources — which is exactly what the citations are there for — and against your local protocols.