About the Project

Inspiration

In an era where AI-generated content is everywhere, I started noticing a growing problem: people are no longer questioning what they read. Content is consumed quickly, shared instantly, and rarely verified. While AI tools make information more accessible, they also blur the boundary between human-authored and machine-generated content.

What inspired this project was a simple question:
What if we could see trust, instead of guessing it?

We already use tools like lenses to examine the world more closely — but when it comes to information, we are still reading blindly. This project began as an attempt to design a “lens” for understanding content.

What I Built

I built TrustLens, an AI-powered tool that analyzes text or URLs and visualizes their trustworthiness through an interactive interface.

Instead of giving a binary answer, TrustLens evaluates content across multiple dimensions:

AI authorship likelihood
Source credibility
Factual accuracy
Logical consistency
Emotional objectivity

These dimensions are combined into an overall trust score:

[ \text{Trust Score} = \frac{w_1 A + w_2 C + w_3 F + w_4 L + w_5 E}{\sum w_i} ]

Where:

( A ): AI authorship score
( C ): credibility
( F ): factual accuracy
( L ): logic
( E ): emotional neutrality

The system presents the result as a “trust label,” along with:

Sentence-level AI probability highlighting
Claim and citation verification
Logical issue detection
Emotional tone analysis
Optional human-style rewriting

How I Built It

TrustLens was built using a combination of modern web technologies and AI models:

Frontend: React + Next.js
Backend: API-based architecture with OpenAI models
Interaction design: real-time feedback with debounce and progressive rendering

The analysis pipeline includes:

Parsing input (text or URL)
Segmenting content into sentences and paragraphs
Running multi-dimensional evaluations using AI models
Aggregating scores and generating explanations
Rendering interactive visual feedback

A key design decision was to move beyond static results and create an experiential interface. Low-trust content is not only labeled but also visually transformed through effects like blur, jitter, and pixelation, allowing users to feel instability rather than just read it.

Challenges

One of the biggest challenges was balancing technical complexity and user understanding.

Ambiguity of AI detection
AI-generated content is not always clearly distinguishable. Instead of forcing a binary output, I designed a probabilistic and explainable system.
Designing for multi-dimensional trust
Trust is not a single metric. Combining multiple evaluations into a coherent and readable system required careful structuring.
Avoiding information overload
Providing rich analysis without overwhelming users required layering information and designing progressive disclosure.
Translating abstraction into interaction
Concepts like “trust” and “uncertainty” are abstract. I explored how visual instability and distortion could communicate these ideas more intuitively.

What I Learned

Through building TrustLens, I realized that designing AI tools is not just about generating results, but about shaping how people interpret information.

I learned the importance of: