Truthlens

Every day, we see information that may not be true. Fake news and deepfakes make trust harder. Truthlens helps detect bias, manipulation, and misinformation making truth easier to find.

Comment

About TruthLens

Inspiration

TruthLens was inspired by one of the biggest problems in today’s digital world: misinformation.

Every day, people consume massive amounts of information through social media, news platforms, videos, and online communities. But not all of it is true. Fake news, emotional clickbait, manipulated images, and AI-generated deepfakes are becoming more common and harder to detect.

As AI becomes more advanced, misinformation is no longer limited to text. It can now appear as realistic videos, cloned voices, and edited visuals that look convincing enough to fool anyone.

This raised one important question:

How can we help people verify what they see before they believe or share it?

That question became the foundation of TruthLens.

What it does

TruthLens is a verification assistant designed to help users detect and understand misinformation.

Users can submit:

  • News articles
  • Social media posts
  • Video links
  • Audio clips

The system analyzes the content through multiple layers:

  • Text Analysis (NLP) to detect emotional manipulation, bias, and misleading language.
  • Deepfake Detection to identify inconsistencies in video frames, lip-sync, and voice patterns.
  • Propagation Analysis to detect suspicious sharing behavior and bot-driven activity.
  • Verification Checklist to guide users through practical fact-checking steps.

TruthLens does not just tell users if something is suspicious. It explains why.

The goal is not only to detect misinformation, but also to strengthen digital literacy.

How we built it

We built TruthLens using a multi-layer AI architecture.

1. Text Analysis

We used transformer-based NLP models inspired by BERT and RoBERTa to identify:

  • clickbait language
  • exaggerated claims
  • unsupported statements
  • emotional triggers
  • manipulative wording

This helps detect content designed to influence emotions rather than provide facts.

2. Deepfake Detection

For video and audio analysis, we explored hybrid models:

$$CNN + LSTM$$

Where:

  • CNN detects frame-level visual anomalies.
  • LSTM detects time-based inconsistencies like flickering or unnatural blinking.

We also used cross-modal alignment:

$$LipSync \leftrightarrow AudioWave$$

to compare speech timing with mouth movement.

This helps detect deepfake videos and manipulated audio.

3. Propagation Analysis

To understand how misinformation spreads, we modeled content as a network:

$$G = (V, E)$$

Where:

  • V represents users or accounts
  • E represents interactions like shares, reposts, and comments

By analyzing how information spreads, TruthLens can identify unnatural patterns such as bot-driven amplification or coordinated mass sharing.

4. Explainable AI

To avoid black-box decisions, we integrated explainability methods.

Instead of simply labeling content as suspicious, TruthLens provides reasons behind the result, helping users understand what triggered the detection.

This makes the system more transparent and trustworthy.

Challenges we faced

One of the biggest challenges was combining multiple detection systems into one smooth experience.

Text analysis, deepfake detection, and propagation analysis all require different processing methods, and integrating them into one workflow was complex.

Another challenge was balancing speed and accuracy. Since this project was developed during a hackathon, we had to focus on building a strong proof of concept while keeping the experience simple and practical.

We also faced the challenge of making the results understandable for normal users, not just technical users.

What we learned

This project taught us that misinformation is not just a technical issue. It is also a human issue.

We learned:

  • how multiple AI systems can work together
  • why explainability is essential in trust-based products
  • how digital literacy matters just as much as detection
  • how user behavior affects product adoption

Most importantly, we learned that helping people access the truth means helping them think critically.

Product Vision

One of the core ideas behind TruthLens is simple:

People should not have to change their habits to verify the truth.

Instead of asking users to open another app every time they want to fact-check something, TruthLens is designed to fit naturally into the way they already consume information.

Our long-term vision is to build TruthLens as a browser extension, similar to Grammarly.

While users browse social media, read articles, or watch videos, TruthLens can quietly analyze content in real time and provide credibility insights instantly.

For example, it can:

  • flag suspicious claims
  • detect manipulative wording
  • warn users about possible deepfakes
  • highlight missing evidence
  • point out biased framing

This makes verification seamless, natural, and part of everyday online behavior.

By integrating directly into user habits, TruthLens lowers friction and increases the likelihood of verification before sharing.

What's next

In the future, we want to expand TruthLens into:

  • a browser extension for real-time fact-checking
  • social media integrations
  • educational tools for schools and universities
  • API services for news platforms and media companies

Our vision is simple:

Make truth easier to access, and misinformation harder to spread.

Built With

  • chatgpt
  • google
  • loveable
  • news
  • openai
Share this project:

Updates

Submission history