From research software to a SaaS platform

Cut hundreds of hours of manual effect-size coding to a few, with the audit trail intact.

M-AIDA reads full-text PDFs with a large language model, proposes the statistics, and lets your team verify, lock and release every record. AI proposes; humans decide. Built for systematic-review and meta-analysis teams in any field.

Open the free demo Talk to the authors

DOI 10.5281/zenodo.21282516 Copyright registration via Can Tho University Open source, auditable 25 pinned conversion tests

Founder · principal investigator Je m’appelle Huong. I built M-AIDA for my own dissertation first. Every claim below is what it already did for me. AI avatar of author Do Thuy Huong waving hello

Product proof, not promises

Battle-tested on a real doctoral meta-analysis.

The P6 study of the founding dissertation used M-AIDA end to end: every number below came out of the locked dataset and feeds a published, DOI-archived analysis.

236

Studies processed

286

Effect sizes locked

100%

Records human-verified

1977-2026

Search window covered

The bottleneck

Screening tools stop where the real cost begins.

Manual coding

Slow and error-prone

Extracting and converting statistics from hundreds of PDFs takes months and quietly accumulates transcription errors that no reviewer can see.

Raw LLM

Fast but untrustworthy

Pasting papers into a chatbot yields numbers with no provenance, hallucination risk, and nothing you can defend in front of a journal or committee.

M-AIDA

AI speed, human authority

The model proposes each statistic with a confidence score; your investigator verifies, overrides and locks it. Only locked records ever leave the system.

What you get today (v7.1.1)

A complete effect-size preparation line.

Extract

PDF to structured record

The LLM reads the full text and proposes N, r, t, df, beta, p and CI with a data year, per record.

Convert

Published formulas only

t and beta convert to r by Cohen (1988) and Peterson & Brown (2005); the formula, inputs and result are stored on every record.

Confidence gates

Nothing slips through

Three-tier confidence (1.0 / 0.8 / 0.6); anything below 0.70 is flagged for mandatory review before it can move on.

Verify

Investigator dashboard

A human inspects every field, overrides what is wrong and approves. The model never assigns moderators or interprets results.

Lock & audit

Immutable by design

Approved records are sealed with a timestamp and cannot be edited; the audit trail satisfies journal AI-disclosure requirements.

Release

Straight into R

Export a plain CSV read directly by metafor; nothing about your analysis is held inside the tool. BYOK keeps you vendor-neutral.

How it works

Four steps from paper to analysis-ready data.

Upload PDFs

Drop the full-text papers of your included studies.

AI proposes

Each statistic arrives with a confidence score and its source.

You verify

Confirm or override every field; flagged records cannot be skipped.

Lock & export

Locked records export to CSV for metafor, with the full audit trail.

Who it serves

Built for every team that codes evidence.

Doctoral researchers

One dissertation, one budget

Run your own meta-analysis end to end on the free tier, exactly the way the founding dissertation did.

University review units

Many projects, many coders

Shared projects and quotas for systematic-review labs across medicine, economics, management, psychology and education.

Pharma / HTA evidence teams

Regulatory-grade trails

Immutable locks and audit logs matched to health-technology-assessment and submission workflows.

Journals & methodologists

Verifiable submissions

Ask authors for the locked dataset and audit trail instead of trusting a spreadsheet.

Plans

Anchored to the time you save, not the tokens you burn.

Free0đ forever

1 project
Monthly extraction quota
Full verify-lock-export flow
Community support

Start free

ProContact

Unlimited projects
Medium extraction quota
Managed LLM included
Email support

Join the pilot

Team / LabPer seat

Multi-coder collaboration
Inter-rater agreement (kappa/ICC)
High quotas, shared projects
Academic discounts

Talk to us

EnterpriseAnnual

SSO · BYOK
On-premise option
Dedicated support & SLA
Security roadmap (SOC 2 / ISO 27001)

Request a call

Transparency note: M-AIDA Cloud is in the pilot phase of the v1.0 commercialization plan (09 Jul 2026). Tier structure follows that plan; official prices will be published after intellectual-property arrangements with Can Tho University are finalized. LLM credits are available on every tier for usage beyond quota. The current open-source v7.1.1 remains free for academic self-hosting.

Roadmap

From today's tool to a full platform, honestly labeled.

Shipped · v7.1.1

Registered reference release

Extract, convert, verify, lock, export. DOI-archived, copyright registration filed, used by the founding dissertation.

Phase 1 · MVP

A sellable cloud (2-4 months)

Accounts, organizations and projects on PostgreSQL; managed LLM with quotas; Stripe billing; generalized beyond international business.

Phase 2 · Teams

Collaboration and integrations (4-8 months)

Multi-coder workflows with kappa/ICC agreement; Zotero, Mendeley and PubMed import; RevMan and CMA export; API and mobile companion.

Phase 3 · Enterprise

Compliance and scale (8-14 months)

SSO, BYOK, on-premise deployments for health and pharma customers, and a SOC 2 / ISO 27001 certification track.

Founding team

Built by the people who needed it most.

Do Thuy Huong

Co-founder · Principal investigator · Ph.D candidate

Designed and built M-AIDA for the P6 meta-analysis of her dissertation on internationalisation and firm performance across Asia; verified and locked all 286 effect sizes herself.

ORCID 0000-0002-7711-2487 · thuyhuongctu@gmail.com

Assoc. Prof. Dr. Phan Anh Tu

Co-founder · Scientific supervisor · Co-author

Associate Professor at the School of Economics, Can Tho University; supervises the founding dissertation and co-authors M-AIDA, anchoring the method in international-business scholarship.

ORCID 0000-0003-0667-3137 · patu@ctu.edu.vn

FAQ

Straight answers.

Does M-AIDA replace Rayyan, Covidence or DistillerSR?

No. Those platforms excel at search and screening. M-AIDA specializes in the stage after them: extracting and converting effect sizes with a human gate, then handing a locked CSV to R. Most teams use both.

Does it run the meta-analysis for me?

Deliberately not. Estimation belongs to established, open methods such as metafor in R. M-AIDA prepares the data so that step is trivially reproducible.

Where does my data live?

Today: on your own machine; v7.1.1 is self-hosted and BYOK, so PDFs go only to the LLM provider you configure. The cloud edition will add encryption, audit logs, GDPR and Vietnam Decree 13/2023 compliance, and a policy of never training on customer data.

Who owns the software?

Authored by Do Thuy Huong and Phan Anh Tu; copyright registration is filed via Can Tho University as co-owner. The commercial license model (dual-license) is being finalized with the university before official pricing.

What are the technical requirements today?

FastAPI backend, React front end, Docker for local deployment, plus an API key for the LLM provider of your choice. The in-browser demo needs nothing at all.