Creator · AI workflow
Bring the evidence back into the conversation.
A workspace for asking whether supplied sources support a claim—and inspecting the passages behind the answer.
ClaimCheck
The problem
When reviewing a claim, the useful question is often specific: what do these sources actually support? That requires clear provenance, readable passages, and room for human judgment.
My contribution
I created ClaimCheck to connect source extraction, passage retrieval, and review in one workspace. It supports a manual ChatGPT handoff and keeps human review separate from model judgments.
How the workflow connects
- Add sources
- Inspect text
- Find passages
- Review claim
Workflow overview—not an application screenshot.
Decisions behind the build
Inspect the source text
Browser extraction and OCR make source text available for review before it becomes evidence. Users can check the extracted text against the original.
Keep passages close
Deterministic retrieval finds candidate passages, and users can select additional evidence rather than relying on an unexplained answer.
Make uncertainty visible
Support, contradiction, and missing evidence are different outcomes. The workspace keeps notes and review decisions distinct from model output.
Where it stands
The public sample uses illustrative fixtures. Live inference is disabled in the documented deployment; the complete private owner workflow remains to be verified. Source support is not proof of real-world truth.
Based on project documentation reviewed October 18, 2026.