Verify Answer
Neural Proof Engine Live SSE Pipeline Active

Check the answer before you trust it: Neural Proof evidence-checking AI assistant

Autonomous evidence checking for LLM responses. Neural Proof deconstructs answers into atomic propositions, audits citations, browses official web sources with TinyFish, and computes grounded trust scores in real time.

Verify an Answer How It Works

Powered by

TinyFish
OpenAI OpenAI
Groq

Interactive Engine

Check the Answer Before You Trust It

Paste any AI-generated response or select a test scenario. Neural Proof extracts verifiable propositions, queries authoritative web sources via TinyFish, and validates quotes with RapidFuzz token matching.

Neural Proof SSE Engine Pipeline Ready • POST /api/verify
● 01
Extract Claims
Groq Llama 3.3
○ 02
Authoritative Search
TinyFish Search
○ 03
Fetch & Grounding
RapidFuzz ≥ 85%
○ 04
Agent Escalation
TinyFish Browser
○ 05
Trust Score
Verdicts & Audit
Live Pipeline Event Stream (Server-Sent Events)
text/event-stream
00:00:00.000 [idle] Awaiting input... Select a scenario or click "Run Neural Verification"
--%
Neural Grounded Trust Score

Run verification to compute grounded trust score.

0 Supported
0 Partial
0 Contradicted
0 Insufficient
Answer Grounding Heatmap Click claim to inspect source

The original AI response will appear here with claims color-coded by verified truthfulness.

Audited Proposition Cards Authoritative Excerpts

About Neural Proof

  • 100% Grounded in Official Web Sources
  • Autonomous Browser Escalation via TinyFish Agent
  • Zero Tolerance for Fabricated Citations & Dead Links
  • Open-Source Llama 3.3 70B Orchestration via Groq
  1. 01

    Proposition Extraction

    Groq Llama 3.3 70B deconstructs responses into atomic, falsifiable propositions with temporal markers.

  2. 02

    Authoritative Search

    TinyFish Search queries official domains (.gov, .edu, who.int, arxiv.org) with temporal awareness.

  3. 03

    RapidFuzz Grounding

    Raw source markdown from TinyFish is verified using partial token ratio ≥ 85% to eliminate hallucinated excerpts.

  4. 04

    Trust Score & Audit

    Dynamic SPAs trigger the TinyFish Agent. A composite trust score, citation audit, and live event stream are returned.

Deployment Models

Open & Scalable
Verification

Open Source

Free MIT License • Open Source

Self-host with your Groq API key and TinyFish standard endpoints.

  • FastAPI SSE Streaming Pipeline
  • Groq Llama 3.3 70B JSON mode
  • TinyFish Search & Fetch API
  • Offline DEMO_MODE included
View Repository

Autonomous Agent

Recommended

Live Real-time verification

Automated compliance and hallucination checks for production LLM apps.

  • TinyFish Browser Agent Escalation
  • RapidFuzz Grounding Audit (≥85%)
  • Citation & DOI Hallucination Check
  • Temporal Outdated Deadlines Detection
  • Sub-3s streaming response time
Try in Workbench

Pipeline SDK

Modular Python & REST

Embed Neural Proof into LangChain, LlamaIndex, or custom agent pipelines.

  • REST & SSE Server-Sent Events
  • Dockerized container deployment
  • Custom domain authority operator
  • Full audit trail JSON exports
API Documentation