GRACIE

Inspiration

Before understanding the insidious cycles of narcissism, gaslighting, love-bombing, and devaluation, I spent years navigating a profoundly toxic environment. This profound personal loss and confusion fueled my unwavering determination to build a tool that could prevent anyone else from enduring similar experiences. My vision was to create the "flashlight in the dark" that was desperately needed during those challenging times.

What it does

GRACIE (Gaslighting Recognition and Cognitive Intelligence Engine) is an AI-powered analytical tool designed to help users identify and understand manipulative patterns in conversations. It functions by allowing users to:

• Upload text screenshots of conversations.
• Paste any conversation text directly.

GRACIE then leverages AI precision to provide a comprehensive analysis, including:

• A clear manipulation severity scale.
• Detailed breakdowns of behavioral patterns, highlighting narcissistic and gaslighting tactics.
• Actionable steps tailored specifically to the user's situation, empowering them to navigate or address the identified manipulation.

COACH is GRACIE's friend. GRACIE will send your results to COACH, whom you can then talk to about the manipulation through chat. I also provide resource information to make sure that you are doing what's best for your safety.

How I built it

GRACIE was built on a robust foundation of natural language processing (NLP) and machine learning. I utilized a custom-trained large language model (LLM) fine-tuned on a proprietary dataset of conversational patterns, specifically curated to identify subtle nuances of gaslighting, emotional manipulation, and narcissistic behaviors. The front-end was developed with React, ensuring a smooth and intuitive user interface for seamless text and image uploads. The backend infrastructure, running on a serverless architecture, provides scalable and secure processing of user inputs, ensuring rapid and accurate analysis.

Challenges I ran into

One of the primary challenges I ran into was curating a diverse and ethically sound dataset for training the AI. Distinguishing genuine emotional expression from manipulative tactics required meticulous annotation and validation. Ensuring the AI's recommendations for "actionable steps" were truly helpful, empathetic, and avoided promoting conflict was another significant hurdle, requiring extensive iterative testing with psychological experts. Finally, handling various images and texts efficiently while maintaining privacy was a technical gauntlet.

Accomplishments that I'm proud of

I am immensely proud of GRACIE's ability to offer a tangible tool for clarity and empowerment to individuals in confusing or toxic situations. Developing an AI that can accurately identify the nuanced patterns of gaslighting, often imperceptible to the untrained eye, is a significant technical and ethical achievement. I am particularly proud of the positive feedback from early testers who found GRACIE's insights profoundly validating and helpful in understanding their own experiences. The seamless integration of text and image analysis into a single, user-friendly platform is also a testament to my dedication.

What I learned

I learned the critical importance of interdisciplinary collaboration (even as a solo developer, understanding the perspectives of psychologists and UX principles was key) to build a truly impactful tool. The journey highlighted the ethical responsibilities inherent in developing AI that analyzes human communication and behavior. I gained deeper insights into the subtle yet devastating effects of psychological manipulation and the immense need for accessible resources to combat it. Furthermore, I refined my understanding of building emotionally intelligent AI that delivers insights not just diagnostically, but also constructively and supportively.

What's next for GRACIE

My immediate plans for GRACIE include expanding its language capabilities to serve a global audience and integrating with popular messaging platforms for even easier conversation analysis. I aim to develop more personalized "actionable step" recommendations through adaptive learning, evolving with user feedback. Further down the line, I envision GRACIE becoming a comprehensive resource for mental well-being, potentially offering connections to verified support networks and educational content on healthy communication. I am committed to continuous refinement of its AI models to stay ahead of evolving manipulative tactics and provide an ever more precise and compassionate tool for its users. Possibly one that can be integrated into your text messaging which will analyze texts as you receive them.

Built With

• gemini
• javascript
• json
• react
• stripe
• supabase
• tailwind
• typescript
• vite