Inspiration
While using AI tools like ChatGPT,etc. I noticed how easily personal information such as names, emails, or locations gets shared without any privacy control. Once sent, users lose visibility and control over their data. This gap motivated me to create Ghost Layer, a simple but effective privacy layer for AI interactions.
What it does
Ghost Layer automatically detects and anonymizes sensitive information in user input before it reaches an AI model. It ensures the AI receives meaningful context while the user’s identity remains protected.
How we built it
I built Ghost Layer using a Chrome extension for input interception and a local Flask backend. Microsoft Presidio and spaCy handle PII detection, while Faker generates realistic replacements. All processing happens locally.
Challenges we ran into
Balancing accurate PII detection with real-time performance and managing reversible mappings without increasing privacy risk were key challenges.
Accomplishments that we're proud of
Balancing accurate PII detection with real-time performance and managing reversible mappings without increasing privacy risk were key challenges.
What we learned
I learned how to design privacy-by-design systems and apply NLP to real-world security problems.
What's next for Ghost Layer
Next, I plan to add encrypted local storage, support for more AI platforms, and mobile phone integration, enabling Ghost Layer to protect AI interactions on smartphones through a lightweight app or system-level keyboard extension
Log in or sign up for Devpost to join the conversation.