AI literature search that answers in plain words, with the passage behind every sentence
Ask the literature in plain words and get a sourced answer in seconds.
See one case, screen by screen ↓A translational scientist wants to know how tumours escape a covalent KRAS G12C inhibitor through MET. The answer is spread across a plasma series in a journal, a cell and xenograft paper, a patent family, an in vivo pharmacology report from Building 2 and signed entries in a Laboratory 4 notebook. Each lives in a different system, and the internal ones are hard to find at all.
So the afternoon goes on searching, reading for the one passage that matters, and checking the figure against the table it came from. And an answer passed on in a meeting without its source cannot be trusted or reused — the next person asks the same question and starts again.
Estimated split of a scientist’s week of searching, by hand and with the solution.
Four specialist agents index, answer, check and watch the literature; a scientist decides what to keep.
Dr. Maya Chen, a translational scientist on Programme CVT-416, needs to know what is known about resistance to KRAS G12C inhibitors through MET amplification. Here is what happens, screen by screen, in the working solution.
Overnight the library agent indexed 1,284 new records, and the watch agent matched them against what Maya asked to be told about. Her home page lists a new 212-patient plasma series, an updated in vivo pharmacology summary for Study RX-0942, and a patent family whose claims 4 and 7 touch the study plan.
“Annals of Molecular Oncology · indexed yesterday · matches your watch “KRAS G12C resistance””
Maya asks the question as she would ask a colleague. It is scoped to Programme CVT-416, across papers, patents, internal reports and lab notebooks she may open, from 2021 to 2026. The steps run in view: 128,412 sources searched, 36 passages read in full, an answer written across 7 sources, 3 of them our own.
“Checked every sentence — 9 sentences, 9 supported by the passage cited, 0 removed”
MET copy gain is the most common single escape route — 23 of 212 people at progression, 10.8% — and the VLT-302 cohort matches at 4 of 38. Plasma shows the gain a median of 46 days before the matched biopsy, and adding a MET inhibitor restored suppression in published models and in RX-0942. Every sentence carries a numbered citation, and the answer is marked high confidence.
“A published family claims the combination at 200–400 mg of the MET inhibitor, which overlaps the 300 mg arm in the current plan.”
The plasma series, the pooled paired plasma and tissue cohorts (389 people), the cell and xenograft models, Study RX-0942 and the VLT-302 cohort — each with its design, size, what it found and what it means. Beside it, the sources used, with internal work marked “Our own”.
Any row or citation opens the source on the page it was read from — here page 3 of 14 of the plasma series, with the sentence on 23 of 212 samples highlighted. The panel says why it came back and who may read it.
“The citation checker read the sentence against this passage and found it fully supported.”
Maya keeps the answer as “Resistance through MET copy gain — what we know today” in Programme CVT-416. The 7 sources, the 9 citations and the five-study evidence table are kept with it, with her name and the time.
The notebook holds what the team decided to keep: plasma at progression as the entry test, the combination restoring suppression, a dosing window that overlaps a published claim, one third of progressions that still show nothing. A programme lead can publish a note to the programme, and the evidence table exports for a report or a review.
Asked whether anyone has dosed a MET inhibitor with a KRAS G12C inhibitor in people, the agent says the library holds no such report. It shows what it does hold — models only, RX-0942 in mice, and a patent dose range that is a claim, not a result — at medium confidence, with “Gap noted”.
“No passage in the library reports this combination given to people. Rather than write a sentence without a source, here is what the library does hold.”
318 questions answered by 46 scientists in the last 30 days, 2,914 sentences checked and 7 cut in draft, 94 answers kept and reused 212 times. 31% of cited passages come from internal research reports.
“Nearly a third of every answer comes from our own research reports — work the company already paid for and could not find before.”
Every sentence must carry a source, confidence shows on every answer, and sources older than 2021 are marked when cited. Internal reports and lab notebooks are open only to their programme team, and every question is logged with the passages returned.
Switching “Sentence must carry a source” off is locked: only the library administrator can change it, and the change is written to the audit trail.
The same library, seen by four people on the programme — what their week looked like, and what it looks like now.
Four specialist agents index, answer, check and watch the literature; a scientist decides what to keep.
Reads each new source, splits it into passages, and tags it by target, mechanism, indication, modality, study design and population size, so answers can be scoped and open the source where it was read.
Takes a question in plain words, searches every source type, and writes an answer in which each sentence carries the passage behind it — with an evidence table comparing the studies.
Reads the drafted answer back against the passages, sentence by sentence, and removes or flags anything the source does not actually say.
Runs each night against what people asked to be told about, ranks new sources by how close they are to the watcher’s programme and recent questions, and writes the morning list with one line on why each matched.
Decides what to keep. The agents propose; a named person decides.
Beside every screen, the assistant answers questions about the library — or takes an instruction about what to watch.
What changed on my watched topics this week?
Three sources matched KRAS G12C resistance since Monday: a 212-patient plasma series, an updated internal pharmacology report, and a published patent family whose claim 4 overlaps your 300 mg arm. The patent is the one worth ten minutes today.
Watch MET bypass resistance in patents too
Done. I widened the MET bypass signalling watch to patents as well as papers, and it will tell you each morning. I have added it to Settings so the team can see what is watched.
Build an evidence table comparing the resistance studies
The evidence table is on your answer: five studies, their design, how many people or models, what each found, and what it means. Every row opens the passage it came from.
Who else in the team asked about this target?
Four people besides you asked about this target in the last week — Sam Patel twice, Lena Ortiz on the assay, Omar Haddad on the patent position and Dana Okafor on regrowth timing. Their questions and the sources they were given are in the notebook.
11 screens from the working solution, on its sample data. Pick one to see it large.
How answers are written, who may see which programme and what is watched are settings, not a project.
| Setting | Default | Choose from |
|---|---|---|
| Sentence must carry a source | On | On · Off — locked to the library administrator |
| Show confidence on every answer | On | On · Off |
| Oldest source to use | Published 2021 or later | Published 2021 or later · Published 2016 or later · Any year |
| Answer length | Standard | Short · Standard · Full review |
| Who sees Programme CVT-416 | Programme team · 24 people | Programme team (24 people) · Programme team + leads (29 people) · Everyone in research |
| Published papers and patents | Everyone in research | Everyone in research · Everyone in the company |
| Questions are logged | On | On · Off |
| What the agents watch | KRAS G12C resistance, each morning | Any topic · each morning or weekly digest |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample research library · “target” = the design goal, measured in the live solution · “estimated” = our estimate from published figures · “typical” = published surveys of scientists’ time (18–39% of a week spent searching). People, companies and products named on this page are fictional — characters and sample data in the working solution.
Asking a question in plain words and getting an answer drawn from the literature, instead of a list of hits to read. Literature Intelligence answers across published papers, patents, internal research reports and signed lab notebooks at once, and every sentence carries the passage it came from.
The research agent writes one claim a sentence and attaches the passage behind it. The citation checker then reads each sentence against its passage, removes anything the source does not say and marks partial support as medium confidence, with the reason.
It tells you. The answer says no passage in the library answers the question, and shows what the library does hold — for example, models only, or a patent claim that is not a clinical result — marked “Gap noted”.
Yes. Internal research reports are indexed on upload and lab notebook entries once they are signed off. They are open only to the programme team that owns them, and answers mark which sources are your own.
Yes. Claim text is read and quoted with the family, the claim number and the date it published. In the working solution an answer flags that claim 4 of a published family covers a dose range overlapping the 300 mg arm in the study plan.
Yes. You set watches on topics — KRAS G12C resistance, MET bypass signalling — over the source types you choose. The watch agent runs each night and writes a morning list or a weekly digest, with one line on why each source matched.
Yes. The agents answer; a scientist decides what to keep in the programme notebook, and a programme lead decides what is published to the programme. Every question is logged with the passages returned.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your sources, programmes and access rules — and it goes live once every quality gate has passed. We will show it to you on your own literature first.
Every rival trial, readout and approval lands in a sourced brief for each programme, every Monday.
1-min filmPreclinical ReportsRaw study data becomes a draft report in hours, with every number traced to its source.
1-min filmScience HarvesterEvery inherited file read and cited, and a signed dossier for each asset before the experts leave.
1-min filmWe’ll run Literature Intelligence on your own papers, patents and research reports.