Evidence synthesis at corpus scale
Lyra: medical research intelligence. Harmony in evidence.
Lyra is our scholarly fiduciary agent. She runs systematic literature search, supports meta-analysis, reviews clinical trials, and identifies research gaps. Every claim ships with the evidence behind it. No fabricated PMIDs, ever.
Try Lyra via API Read the Lyra deep dive95M+ research vectors indexed · Citation-precise · CEBM evidence grading · Corpus auditable
What Lyra is built for
Methodical, thorough, citation-precise. Like a brilliant research librarian who knows every paper and every limitation.
Systematic literature search
Run structured searches across 95M+ vectors of medical research. Lyra returns ranked evidence with collection of origin, study type, and the rationale for each retrieval. PRISMA-friendly output on request.
Meta-analysis support
Identify candidate trials for a meta-analysis, surface effect sizes and confidence intervals where reported, flag heterogeneity and risk-of-bias considerations across the included studies.
Clinical trial review
Summarize a trial's design, population, primary and secondary endpoints, statistical analysis plan, and limitations. Lyra grades evidence using Oxford / CEBM levels and labels uncertainty plainly.
Research gap identification
Given a clinical question, Lyra surfaces what the literature does and does not currently support, and where the next study could productively land. Useful for grant scoping and protocol design.
Citation-precise answering
Every claim Lyra makes carries a citation back to the underlying paper (PMID where available, collection name otherwise). If a PMID cannot be verified, Lyra refuses to invent one and says so.
Contradiction flagging
When trials disagree, Lyra surfaces the disagreement rather than picking a side. She labels heterogeneity, conflicting findings, and insufficient evidence honestly.
Corpus access
Lyra has prioritized access to the platform's research collections. Every retrieval is auditable to the collection of origin.
| Collection | Vectors | Lyra Boost | Source Type |
|---|---|---|---|
pubmed_abstracts |
5.1M | 1.5x | Peer-reviewed biomedical abstracts |
openalex_top_cited |
16.5M | 1.4x | Top-cited scholarly works (OpenAlex) |
pmc_fulltext_cited |
5.2M | 1.4x | PMC open-access full-text papers |
pkg2_cited_structured |
13.1M | 1.3x | Citation-structured scholarly graph |
pkg2_keywords |
7M | 1.2x | Keyword-indexed research |
pkg2_abstracts |
48M | via Asha | Extended abstracts corpus |
statpearls_clinical |
76K | 1.3x | Clinical reference articles |
clinical_guidelines |
varies | 1.2x | Society and government guidelines |
arxiv_multidomain |
varies | 1.1x | Cross-domain preprints |
Boost values from agents/lyra/agent.yaml. Asha (with mode: all) can also query the full 829-collection platform corpus on demand and hand results to Lyra for synthesis.
Why researchers choose DNAi
Three commitments that change what AI is allowed to do in research workflows.
Citation precision
Every answer ships with the evidence behind it. Collection of origin, PMID where available, evidence grade where applicable. You can verify every claim before you cite it in your own work.
Corpus auditability
The retrieval logs the collections queried, the rank of each result, and the score that promoted it. Methodologists can reproduce a Lyra search by running the same query against the same collections.
No fabricated PMIDs
Sacred refusal #9 in the medical agent contract (canonical list in
core/prompts.py): never fabricate PMIDs or citation
numbers. If Lyra cannot find an actual paper for a claim, she will
say "I do not have evidence for this" rather than make one up.
Sample researcher use cases
Concrete query patterns Lyra excels at. Each one is a real workflow the agent has been tuned for.
Pre-protocol scoping
"Summarize the state of evidence on intermittent fasting and HbA1c reduction in adults with prediabetes. Include effect sizes, sample sizes, and the largest registered ongoing trials." Lyra returns ranked evidence with PMIDs and trial IDs, flags heterogeneity, and lists open questions for protocol design.
Systematic review triage
"For a systematic review of CBT for adolescent depression (ages 12-18), give me candidate RCTs from 2014-2026 with sample size, randomization quality, and primary outcome measure." Output is ready for PRISMA inclusion screening.
Conflicting-trial reconciliation
"Two recent RCTs on SGLT2 inhibitors and heart failure with preserved ejection fraction disagree on mortality benefit. Summarize the design differences and why the conclusions diverge." Lyra surfaces both trials, the methodological deltas, and the residual uncertainty.
Guideline + evidence gap audit
"Compare current ACC/AHA hypertension guidelines against the most recent five trials. Where does the guideline lag the evidence, and where does it correctly hold the line?" Lyra cross-references the guideline text with PubMed and openalex_top_cited.