PrivatePrompt

• 

  Main Chat Interface

Inspiration

Private Information gets leaked into chatbots all the time over the news which is a serious concern for privacy protection.

What it does

PrivatePrompt is a browser-based gateway that scans user prompts for sensitive entities (regex-based PII detection), replaces them with stable placeholder tokens ([EMAIL_1], etc.), and computes a bounded risk score with an explicit clean / masked / blocked classification. Before any model sees text, enforcement builds only a SafeLlmRequest: downstream calls use maskedText (never the raw prompt) when risk score is below the threshold.

How we built it

Single page React web app and wired it to a prover that just scans and classifies sensitive info based on regex and pattern detection, to calculate a risk score which determines if a query needs to be masked before it goes through AI chatbot

Challenges we ran into

1. Blockchain is a very new concept for us. We have never programmed in it earlier. Initial ideation phase was a challenge as we had to brainstorm to come up with real world use case and how blockchain could be implemented in it.
2. When differentiating which prompts to mask and block, we were unable to a set a practical threshold as a result all of our prompts were getting blocked.
3. Debugging the entire project and understanding the regex expressions was a hassle because some sensitive information was not being detected right. ## Accomplishments that we're proud of

What we learned

1. What is Blockchain and how it can be used to protect privacy
2. Using Typescript and Debugging
3. Communication skills and presentation skills
4. Brainstorming and research analysis to figure out real world problems.

What's next for PrivatePrompt

Use ML models instead of fixed Regex patterns for detection and Integrate actual midnight functionality.

Built With

• openai
• react
• typescript
• vite