IASRD Software · Research Integrity

TRACE-AI

Evidence-gated research analysis. Every finding traced back to verified source text — nothing asserted without it.

Community Edition v1.0-beta External Review Release

Golden Rule: No Evidence = No Claim.

View on GitHub Run it locally

What TRACE-AI does

TRACE-AI is a research-integrity engine, not a summariser and not a chatbot. Upload a document — PDF, Word, CSV, Excel, or a web link — and TRACE-AI checks it against a set of themes, returning only findings that are traceable to a verified passage of source text, each with an explainable confidence score.

The verification engine is fully deterministic: retrieval, scoring, contradiction detection, and status are all rule-based and reproducible. An AI model is used for exactly one thing — proposing clean, readable theme names — and is never permitted to select evidence, assign confidence, or decide a finding's status.

What's included in Community

Deterministic verification

Retrieval, scoring, adversarial review, and contradiction detection — rule-based, reproducible, fully open.

Multi-format input

PDF, Word, plain text, CSV, Excel, and web URLs — pooled into one combined evidence set.

Automatic theme discovery

Themes emerge from the document(s) on their own; manual themes are also supported.

Six-status findings

Verified, Partially Supported, Weak Evidence, Insufficient Evidence, Blocked, Conflicting Evidence — each with a full audit trail.

Evidence hygiene

Metadata, chart noise, and OCR fragments are filtered out before they can become a finding.

PDF highlight viewer

Click a finding to see the exact quote highlighted on its source page.

Exports

JSON, Excel, and Word, each carrying the finding's full audit trail.

See it in action

Screenshots and a short walkthrough video are in progress and will be added here. In the meantime, the README on GitHub documents the full workflow with example output.

Live hosted demo coming soon — after the current stability review. For now, TRACE-AI Community runs locally on your own machine (see below).

Run it locally

TRACE-AI Community is open source under the Apache License 2.0. You can run it on your own machine in a few minutes.

git clone https://github.com/sadiarehman2816/trace-ai.git
cd trace-ai
pip3 install -r requirements.txt
streamlit run app.py

This opens TRACE-AI at http://localhost:8501. By default it runs fully offline and deterministic. Setting an ANTHROPIC_API_KEY environment variable enables HYBRID mode, where an AI model is used only to generate clean theme names — never to select evidence or decide a finding's status.

Full setup notes are in the project README.

About this release

TRACE-AI Community v1.0-beta is an external review release, following a dedicated validation phase against real-world report styles. It is intentionally not labelled a final v1.0 — its purpose is to gather feedback from reviewers and early adopters before that label is earned.

A future TRACE-AI Professional edition is planned as a separate, additive layer for organisation-facing capabilities. Nothing in the Community engine is held back for it, and no licensing changes apply to this release.

Developed by IASRD

TRACE-AI is developed under the International Association for Social Research Dynamics (IASRD), UK Registered Charity No. 1217292, by Sadia Rehman and Muhammad Amin ur Rehman Khan.

How to cite TRACE-AI →