How it works

Explained in
plain English.

You upload research papers. You tell it what to look for. It reads every paper and gives you a clean spreadsheet. Here’s exactly how.

1 Build your form
2 AI reads every paper
3 Review and export
The problem

200 papers. 8 questions each.
1,600 cells to fill by hand.

Most research teams do this manually — reading each paper, highlighting the relevant section, then copying the answer into a spreadsheet. One at a time. It takes weeks. It is exhausting. And it is full of small mistakes that compound into big ones.

0
data points entered by hand
in a typical 200-paper systematic review

1

Tell it what questions
to answer from each paper

Think of it like building a checklist. What do you need to know from every research paper? The age range of patients? What treatment was tested? Whether it worked? You just type your questions in — one by one. No coding. No technical knowledge required.

You can add as many questions as you need, in any structure, and change them at any time.

Extraction Form
3 fields
Patient age grouptext
Treatment receivedlist
Success ratenumber
Generate AI Pipeline
What happens in the background

When you click “Generate AI Pipeline,” two things happen automatically

Step A — Decomposition
It organizes your questions into a work plan

Your questions aren’t all equal — some are independent, some need earlier answers first. The AI groups related questions into extraction tasks, then arranges them into stages. Independent tasks run in parallel (at the same time). Dependent tasks run sequentially (in order). The result is an optimized pipeline plan.

Analyzing your form…
Stage 1Parallel
IdentifyStudyMetadata2 fields
ExtractInterventions2 fields
Stage 2Sequential
ExtractOutcomesdepends on Stage 1
3 tasks · 2 stages · ready to generate code
Step B — Code Generation
It writes real code for every extraction task

From the pipeline plan, eviStreams generates actual Python code — one specialist class per extraction task. Each class knows exactly what fields to find and how. When you upload a paper, these classes execute in stage order and merge their results. You never write a line of code.

Generating code…
class IdentifyStudyMetadata(dspy.Signature):
markdown_content = dspy.InputField()
first_author = dspy.OutputField()
population_code = dspy.OutputField()
class ExtractOutcomes(dspy.Signature):
success_rate = dspy.OutputField()
// depends on Stage 1 results
✓ 3 classes generated · code ready2.4s
Your questionsDecompositionCode GenerationCustom pipeline ready

This all takes a few seconds. Then you upload your papers and it runs.


2

Upload your papers.
All of them.

Drag and drop your PDFs — ten papers or five hundred, it makes no difference. The custom pipeline eviStreams built for your questions now runs on every single paper automatically. You do not read a single word.

< 1sper paper on average — versus 20 minutes by hand
Pipeline running
Processing papers…
paper_001.pdfcomplete0.7s
paper_002.pdfcomplete0.9s
paper_003.pdfcomplete0.6s
paper_004.pdfrunning…
All 127 papers complete
What you get back

Here’s what the computer found

Once the pipeline finishes, eviStreams shows you a table of everything it extracted — one row per paper, one column per question. Every cell has a value pulled directly from the source text. This is your raw data, ready to check.

PaperPatient age groupTreatment receivedSuccess rateStatus
Smith et al. 2021J. Oral Implantology
Adults 45–65Dental implant surgery87%Extracted
Chen & Liu 2020Clin. Oral Res.
Seniors 60+Antibiotic therapy72%Extracted
Kumar et al. 2022Int. J. Oral Sci.
Adults 30–50Bone graft + implant91%Extracted
Patel & Wong 2019Periodontology
Mixed 25–70Medication onlyNot reportedReview needed
Every cell is linked back to its source. Click any value and eviStreams shows you the exact sentence in the original paper it came from — so you can always verify the answer before you accept it.

3

Review what it found.
Fix anything. Download.

You and your team go through each answer. Most will be exactly right. Some you’ll want to adjust. When two reviewers disagree on an answer, eviStreams flags the conflict so you can resolve it together — then you download everything as a spreadsheet, ready for Excel or any research tool.

Nothing is saved until you approve it. You are always in control of the final data.

Review Results
3 fields
Patient age groupAgreed
Adults 45–65, dental implant patients
Treatment receivedConflictAgreed
AI found
Surgery
Reviewer
Medication
Resolve →
Surgery
Success rateAgreed
87% implant survival at 12 months
Export as CSV ↓
The difference
Before eviStreams
×20 minutes per paper, reading and copying by hand
×Copy-paste errors that quietly corrupt your dataset
×One researcher can process roughly 3 papers a day
With eviStreams
Under one second per paper, processed automatically
You review every answer before it touches your dataset
Hundreds of papers processed while you sleep

Still have questions?

Plain answers to the things people ask most.

Yes — if you can write an email and upload a file, you can use eviStreams. There is no coding involved, no technical setup, and nothing to install. Everything happens in your web browser.
You review every answer before anything is saved. eviStreams shows you what it found, and you decide whether it's right. Think of the AI as a very fast first reader — you make the final call on every field.
Any PDF research paper. Medical studies, clinical trials, dental and oral health research, systematic reviews, public health studies — if it's a PDF, eviStreams can read it.
As many as you need. eviStreams processes papers in parallel, so 10 papers takes roughly the same time as 200.
See our Security page for full details. In short: your papers and data are stored securely and are never shared. Projects are isolated — no one outside your team can see your work.
No. eviStreams runs entirely in your web browser. Open the page, log in, and start working — nothing to download or install.

Ready to try it?

Takes about 5 minutes to get started.

Create a free account →

Used by researchers at Penn Dental Medicine, University of Pennsylvania.