Inside Roadrunner

From question to cited synthesis

One workspace for every step of the research process- search, analyze, extract and synthesize

01Deep search

One query, every source

Frame your research question in PICO — population, intervention, comparison, outcome — and Roadrunner turns it into a structured, reproducible query that runs across every source in a single pass. Every query is saved and versioned — your search strategy is a method, not a memory.

  • Boolean and semantic search over titles, abstracts, and full text
  • Detailed, citation-backed summaries for every result
  • Saved, re-runnable queries — export the exact search string for your methods section

A search strategy that’s reproducible by design — saved, versioned, and yours to re-run.

app.roadrunner.ai/…/RQ-042/searchReproducible query
search · RQ-042 · query v35 sources
Roadrunner AI — Gather step: an auto-generated, reproducible database query with ranked search results.
02Screen at scale

Your criteria, at scale

Define inclusion and exclusion criteria once, then screen thousands of records against them. Roadrunner proposes each decision with its reasoning; you confirm, override, or send to review — and every call lands in the audit log.

  • Title/abstract and full-text passes with the same criteria set
  • Borderline records routed to human review, never silently dropped
  • Who decided, when, and why — logged for every record

2,435 records screened in the example review · every decision on record

app.roadrunner.ai/…/RQ-042/screenEvery decision logged
screen · RQ-042 · title / abstract2,435 screened
Roadrunner AI — Screen step: each paper judged against inclusion criteria with a confidence score and rationale.
03Structured extraction

Every cell knows where it came from

Define the fields you need — endpoints, populations, effect sizes, dosing — and Roadrunner fills the table from the full text. Each extracted value links to the exact source passage, down to the page.

  • Cell-level provenance — click any value to open the passage behind it
  • Confidence scores on every cell; low-confidence values flagged for review
  • Custom schemas, reusable across reviews — export to CSV or your stats package

Low-confidence cells are flagged, never guessed — a human signs off before they ship

app.roadrunner.ai/…/RQ-042/extraction/outcomesCell-level provenance
extract · RQ-042 · outcomes4 studies
Roadrunner AI — Extract step: structured outcome data pulled from each study with cell-level provenance.
04Systematic review & synthesis

A PRISMA that draws itself

Identification counts, screening decisions, and exclusion reasons roll up into a live PRISMA 2020 diagram — no manual tallying, no numbers drifting out of sync with the review.

  • Counts update live as screening decisions change
  • Exclusion reasons itemized at every stage, straight from the log
  • Export publication-ready figures and the numbers behind them

Built from your real screening decisions — not a template you fill in later

app.roadrunner.ai/…/RQ-042/prisma-2020Auto-generated
synthesize · RQ-042 · PRISMA 2020live
Roadrunner AI — Synthesize step: a cited systematic-review report generated from the included evidence.
05Chat with full-text papers

Answers with receipts

Ask plain-language questions across your entire full-text library. Answers are grounded in the PDFs you've collected, quote the passages they rely on, and cite page numbers you can open in one click.

  • Cross-paper questions: subgroups, dosing, adverse events, endpoints
  • Direct quotes with page references — verify in one click
  • "Not reported" when the papers don't say — no confident filler

Grounded in your full-text library — not the open web

app.roadrunner.ai/…/RQ-042/chatGrounded answers
chat · RQ-042 · 12 full textsgrounded
Roadrunner AI — a direct, cited answer grounded in the retrieved sources.
The wider toolkit

Analyze papers

Purpose-built capabilities around the core workflow — citation finding, gap analysis, hypothesis comparison, and more — each grounded in the same cited evidence base.

High-impact citation finding

Surface the most-cited, most load-bearing papers behind any claim — so your references carry weight.

citation graph · impact

Research-gap identification

See where the evidence thins out — populations, comparators, and endpoints no trial has answered yet.

coverage map · gaps

Hypothesis comparison

Line up competing hypotheses against the same evidence base and see, claim by claim, which one holds.

claim-by-claim

Literature-review tables

Study characteristics and outcome tables assembled in one pass — export-ready for your appendix.

export-ready

Data & text extraction

Pull endpoints, populations, and effect sizes out of PDFs into structured fields — with provenance intact.

provenance intact

Models tuned for biomedical literature

Tuned to read the clinical literature — trial design, endpoints, and statistics, not the open web.

domain-tuned
Why it holds up

Every claim, on the record

Answers you can source, decisions you can trace, and workflows anyone can rerun.

Evidence-grounded

Every answer is built from sources you can open — never from model memory.

Traceable

Every claim, value, and decision links back to the exact source passage.

Reproducible

Queries, criteria, and flows are saved and versioned, so anyone can re-run them.

Get started

Put your next question on the record

Every stage, from first search to a cited synthesis — with the receipts to back it up.

Free 2-week trial, all features Team and enterprise plans available