TruthLayer

The Problem

Large Language Models can produce answers that sound confident and convincing while containing incorrect, outdated, or unsupported information. This creates a trust problem: users often see a fluent AI response as reliable without knowing which individual claims are actually supported by evidence.

Existing AI assistants may provide citations, but users still have to manually determine whether those sources actually support each statement.

Our Solution

TruthLayer is an AI-powered mobile application designed to act as a verification layer for AI-generated information.

Instead of treating an entire AI response as simply "correct" or "incorrect," TruthLayer breaks the response into individual factual claims and verifies them independently.

The workflow is:

AI Response → Claim Extraction → Evidence Retrieval → Claim Verification → Confidence Scoring → Transparent Results

Each claim is classified as:

Verified — reliable evidence supports the claim.
False — reliable evidence contradicts the claim.
Unverifiable — there is not enough reliable evidence to confidently determine the truth.

For every claim, TruthLayer provides supporting evidence, source attribution, and a confidence score so users can understand not only the result, but also why the system reached that conclusion.

How It Works

A user can provide an AI-generated response by pasting text, using the mobile Share-to-Verify workflow, or providing a screenshot.

The system then:

Extracts individual factual claims from the response.
Generates relevant search queries.
Retrieves evidence from relevant and trusted sources.
Compares each claim against the retrieved evidence.
Detects supporting, contradicting, or insufficient evidence.
Assigns a confidence score to each claim.
Generates an overall Trust Score for the response.
Presents the findings through an easy-to-understand mobile interface.

The goal is not to replace the LLM. Instead, TruthLayer works as a trust and verification layer around AI-generated information.

What Inspired Us

As AI assistants become part of everyday research, education, software development, and decision-making, the ability to distinguish reliable information from confident misinformation becomes increasingly important.

We wanted to explore a simple question:

What if we could see exactly which parts of an AI answer we can trust?

That idea became TruthLayer.

What We Are Learning

Through this project, we are exploring claim extraction, LLM agents, retrieval-augmented generation, evidence grounding, source credibility, confidence scoring, and mobile application development.

We are also studying an important challenge in AI verification: uncertainty. A good fact-checking system should not force every claim into "true" or "false." When evidence is insufficient or sources disagree, the system should communicate that uncertainty clearly.

Challenges

One of the biggest challenges is making verification reliable rather than simply asking another LLM whether an answer is correct.

TruthLayer therefore separates claim extraction, evidence retrieval, source evaluation, and verification into different stages. This allows the system to show users the evidence behind each decision instead of relying solely on the model's internal knowledge.

Another challenge is handling conflicting, outdated, or incomplete information. Our approach is designed to preserve these uncertainties and communicate them transparently.

Our Vision

We envision TruthLayer becoming a reusable trust layer for the AI ecosystem — available as a mobile application today and potentially as a browser extension, API, or middleware layer for AI agents in the future.

AI can generate the answer. TruthLayer helps you decide whether to trust it.