Claim extract
Pull atomic claims from generated text.
AI/ML
Fact-checker Tool
Modular pipeline that extracts claims from LLM-generated text, retrieves supporting evidence, and verifies accuracy using LLM-based checks and tools like Factool. It achieves an F1-score of 0.87 and tackles misinformation with an intuitive UI and transparent analysis reports.
Gallery
Evidence
Retrieve supporting or contradicting sources.
Report
Transparent analysis for each verification pass.
Product
A modular fact-checking pipeline for LLM output: extract claims, retrieve evidence, verify with LLM checks and Factool, then surface transparent reports so users can see why something was flagged.
Problem
LLM-generated text can sound authoritative while carrying unverified claims.
Solution
A staged pipeline with measurable quality (F1 0.87) and an UI that explains verification steps.
Highlights
- Claim extraction → evidence retrieval → verification pipeline
- F1-score of 0.87
- Transparent analysis reports
Specs
- Pipeline
- Extract → retrieve → verify
- Quality
- 0.87 F1
- Tools
- LLMs, Factool
- Language
- Python
- UX
- Analysis reports UI
How it works
- 1Paste or stream LLM-generated text into the tool.
- 2Review extracted claims before verification runs.
- 3Inspect evidence and model judgments per claim.
- 4Export or share the transparent analysis report.
Achievements
- 0.87 F1 on verification quality
- End-to-end modular pipeline with Factool integration
- UI designed for trust—not black-box scores
Stack
PythonLLMsFactool
Core StackFrontendBackend & DatabasesData & Machine LearningDesign & DX