ChurchillonData/claimify-esg

Claimify ESG is a thoughtful and unusually concrete prototype for ESG claim verification. The strongest part is that it does not stop at a polished frontend: the repo includes a real Python pipeline for ingestion, claim extraction, retrieval, reranking, LLM scoring, rationale generation, and a React/Vite interface that exposes the output through Claims, Compare, and Tracker views. The README gives a clear mental model of the system, and the code organization under src/claimify/ is easy to follow. I especially liked that retrieval and scoring persist JSONL outputs, that compute_metrics.py compares predictions against a hand-labelled 60-claim eval set, and that the frontend surfaces methodology details such as source credibility weights, materiality weighting, and evidence trails instead of hiding the model’s judgment behind a single score.

The project also makes good product choices for a sensitive domain. Greenwashing detection can easily become vague or overconfident, but llm_scorer.py uses a strict classification prompt with a high bar for “contradicted,” which is a sensible safeguard against treating general sector criticism as proof against a specific claim. The curated NGO source config, company-specific evidence mapping, and source weighting system make the prototype feel inspectable rather than magical.

The main thing I would improve is reproducibility and trust signaling. GitHub currently reports no detected license, likely because the file is named LICENSE (MIT) rather than the conventional LICENSE; renaming it would help. I’d also add CI for at least Ruff and the lightweight eval scripts, plus a short “methodology limitations” section covering dataset size, source selection bias, LLM dependence, and what the 76.7% eval accuracy means in practice. Since this repo has no stars, forks, issues, or PRs yet and was created recently, a contribution guide, screenshots or hosted demo link, and a documented expected output sample would make it much easier for outside users to adopt or critique. Overall, this is a well-scoped, serious prototype with a strong transparency instinct and a clear path toward becoming a more credible open-source ESG analysis tool.