Skip to content
← Back to projects GitHub

App walkthrough

Claim Inspector

Hybrid RAG + LLM pipeline for source-backed fact-checking

Claim Inspector is not currently hosted online to keep API costs manageable. Scroll through each step below — screenshots on the left, descriptions on the right. Click any screenshot to view it full size.

01

Login

A simple sign-in screen protects the deployment. Authentication keeps the self-hosted instance private while the fact-checking pipeline runs on the server.

02

Dashboard & input

The home view tracks all-time and session spend across Claude Haiku, Voyage embeddings, and billable APIs, with daily usage bars for each service. Below that, paste text or drag and drop a file to start a new fact-check.

03

History & cost estimates

Past runs are saved with timestamps, per-run cost, and flagged-claim counts. You can rename or delete entries. Before analyzing, the app estimates how many claims it will find and what the run will cost.

04

Analysis results

After analysis, each sentence in the source text is highlighted by risk level: verified, medium risk, high risk, or unverifiable. A summary bar shows the breakdown, and optional batch mode trades speed for a 50% discount on Haiku calls.

05

Claim breakdown & sources

Expanding a claim shows Haiku's reasoning against retrieved passages, with direct links to Wikipedia and OpenAlex sources. Each source includes a snippet from the indexed text so you can verify the evidence yourself.

← Back to projects View on GitHub