EvidAI Is Live, The Evidence Platform From Search to Submission
Pull-quote: “In evidence synthesis, the AI’s job is not to be right. It is to be checkable. Every AI decision in EvidAI carries its model, its prompt version, its confidence, and the human who confirmed it.”
The announcement
EvidAI, Zorost’s evidence-synthesis platform for pharmaceutical and research teams, is live in production at evidai.ai. As of this week the platform searches eighteen connected literature sources in parallel and carries a systematic review from protocol to manuscript inside one governed workflow.
Systematic literature reviews are among the most rigorous, and most punishing, workflows in science: months of searching, screening, extraction, and appraisal under methodology standards that leave no room for shortcuts. EvidAI’s position is simple: automate the labor, never the judgment.
What is live today
Search that casts a wide net and shows its work. Eighteen connected sources searched in parallel, including PubMed, OpenAlex, Semantic Scholar, Europe PMC, Crossref, ClinicalTrials.gov, bioRxiv, and medRxiv, with aggregation, deduplication, and per-record source provenance feeding directly into PRISMA reporting. AI-assisted query generation validates MeSH terms against the live NLM headings.
Screening with humans in charge. Dual-reviewer screening is enforced: two distinct human identities, conflicts surfaced rather than auto-resolved, and AI explicitly excluded from the human quorum. The AI assist layer scores every record, and where it is uncertain, a cross-family multi-model consensus reviews the record rather than a single model guessing. Active learning prioritizes the queue with statistical stopping rules.
Appraisal by the book. Risk-of-bias assessment ships with the verbatim RoB 2, ROBINS-I, and QUADAS-2 instruments, and GRADE certainty ratings are computed by a deterministic engine implementing the published method: five downgrade domains, three upgrade factors, arithmetic in code rather than model prose.
Reporting that reconciles by construction. PRISMA 2020 flow diagrams are generated from the actual record counts, and the platform enforces that identified, screened, assessed, and included counts reconcile across every export: DOCX, PDF, LaTeX, CSV, JSON, and a full reproducibility bundle.
Living reviews. Scheduled surveillance re-runs the search and proposes new records under a propose-then-approve workflow; nothing enters the review without an owner’s sign-off.
Every AI action on the platform writes provenance: which model, which prompt version, what input, what confidence, and which human confirmed it, into an append-only audit trail.
Who it is for
Evidence teams in pharma, medical affairs, HEOR groups, and academic centers who need PRISMA 2020-aligned rigor without the eighteen-month calendar. It runs with cloud models or fully local ones for teams whose data cannot leave their walls.
EvidAI is live at evidai.ai. For a walkthrough with your own protocol, reach us through zorost.com/contact.