Open • Transparent • Auditable

AI‑assisted, continuously updated IPCC evidence base

Automates discovery → relevance filtering → chapter assignment → key‑insight extraction → validator checks → live HTML insertion. Built for coverage, provenance, and policy‑readiness.

Why general chatbots such as GPT aren't enough

Fast per‑query answers, but: low coverage, opaque retrieval, no provenance guarantees, and hallucinations in long‑form synthesis. Not fit for assessments.

What specialized tools still miss

Tools such as Elicit, Research Rabbit etc are NOT open-acess, are paid, limited scale, and weak support for consensus building & living updates. We go "glass‑box" + continuous.

How This Tool Improves Upon Existing AI

Our transparent, open-source solution:

Glass Box: Not a black box - fully auditable and transparent
Open Source: Community-driven development and verification
Accessible: Designed for researchers and policymakers
Living Evidence: Continuous updates with versioned HTML reports
Source Tracking: Sentence-level source attribution and validator scores

Pipeline: Citation-graph expansion, semantic retrieval, LLM-based classification, dual-model claim extraction with cross-validation, and human review escalation for low-confidence outputs.

AIPCC Concept Note

Executive summary and concept overview for the AI-augmented IPCC literature synthesis project.

Workflow Diagram (interactive)

Literature Collector - Concept

Chapter assignment — Results

These are the >3000 papers which cite the IPCC report 2021 IPCC_AR6_WGI. We automatically retrieved the papers using literature collector tool shown above. Then we ran ML classification to place them within the existing IPCC chapters. The figure shows the chapter assignments distributions and scores.

Future Directions & Upcoming IPCC Reports

We will use the IPCC AR7 outline (already drafted) to construct a seed set of relevant papers and feed them to literature collector.

Chapter 8: Services & Demand