Bias Mirror

An AI-powered mirror that reveals how perspective and bias shape our thinking.

Comment

Inspiration

I was inspired by how easily conversations today turn polarized, not necessarily because people are wrong, but because they are speaking from different perspectives without realizing it. I noticed that many disagreements stem from framing, hidden assumptions, and cognitive bias rather than pure facts. I wanted to explore whether an AI system could help people see their own thinking more clearly, without judging or correcting them. I wanted to explore whether an AI system could help people see their own thinking more clearly, without judging or correcting them, just like a true friend would!

What it does

Bias Mirror is an interactive Gemini-powered application that takes a user’s belief or opinion and analyzes it through multiple lenses. For a single input, the system:

  • Rewrites the statement in a neutral, emotionally balanced form
  • Reframes it from multiple perspectives (individual, authority, societal, ethical)
  • Identifies implicit assumptions
  • Generates a respectful counter-perspective (“bias mirror”)
  • Tags relevant cognitive biases and framing effects

The goal is reflection, not debate, helping users recognize how perspective shapes interpretation.

How I built it

The project was built using Google Gemini 3 via Google AI Studio, which allowed rapid prototyping without deployment overhead. I designed a structured prompt that leverages Gemini’s long-context reasoning and instruction-following capabilities to produce consistent, multi-section outputs.

The prompt enforces a fixed JSON structure, ensuring the analysis is reliable and interpretable. A lightweight web interface demonstrates how such a system could be integrated into a full application, while the AI Studio app serves as the primary interactive demo for judges.

Challenges I ran into

The main challenge I faced was working with TypeScript for the first time while building the web interface. Adapting to type safety, strict checks, and unfamiliar compiler errors initially slowed development.

However, this also helped me write more predictable and maintainable code. Once I became comfortable with the basics, TypeScript significantly reduced runtime bugs and improved overall confidence in the codebase.

Aside from this learning curve, the project progressed smoothly due to clear scoping and a prompt-first design approach.

Accomplishments that we're proud of

  • Successfully designed a structured, multi-perspective reasoning system using Gemini 3
  • Built an interactive AI Studio app that is publicly accessible and demo-ready
  • Achieved consistent, neutral, and non-judgmental outputs through careful prompt refinement
  • Created a complete end-to-end prototype as a solo developer
  • Designed an AI experience focused on reflection and understanding rather than debate

What I learned

This project highlighted how prompt engineering is a form of system design. Small changes in wording significantly affected tone, neutrality, and reasoning depth.

I also learned how powerful Gemini’s long-context and instruction-following capabilities can be when paired with structured output constraints. Beyond technical skills, the project reinforced the importance of designing AI systems that are human-centered, ethical, and emotionally aware.

What's next for Bias Mirror

Future iterations of Bias Mirror could include:

  • Interactive perspective toggles and visual highlighting of framing shifts
  • Support for longer, multi-paragraph inputs and discussions
  • Accessibility-focused explanation modes for different audiences
  • Optional comparison views to track how opinions evolve over time

The long-term vision is to explore how reflective AI tools can support healthier conversations and better decision-making.

Built With

Share this project:

Updates