AI systems can generate confident answers without clearly showing whether the evidence supports them.

I started ProofPath AI to explore a different approach.

The goal is to help users examine claims alongside supporting evidence and identify gaps in the available information.

Core concept

Claim → Relevant evidence → Support assessment → Overclaim warning → Missing evidence.

Development roadmap

The project proposes:

  • Sentence-BERT semantic retrieval
  • Natural language inference
  • PDF document uploads
  • Visual evidence graphs
  • Source-quality indicators
  • Evaluation benchmarks

The repository documents the concept and an application entry point, but these roadmap capabilities should not be treated as completed or validated.

Important distinction

ProofPath AI is not a universal truth detector. Its intended purpose is to analyse how well user-supplied evidence supports a particular claim.

GitHub repository: https://github.com/tasnem-tech/proofpath-ai

I'm interested in building AI tools that make evidence easier to inspect rather than hiding it behind a single score.