Inspiration
A lot f time need to use regex before asking chatgpt
What it does
It basicallt sanitize the prompt without losing context and it forms back the answer accordingly
How we built it
TMI Prompt Buster uses GLiNER2 zero-shot entity recognition rather than using Regex.
Challenges we ran into
Testing models until found GLiNER2
Accomplishments that we're proud of
Can run locally , cpu only
What we learned
It is much easier to develop when using codex
What's next for TMI Prompt Buster
The next idea is an implicit leak buster: detect details that accumulate over time and identify the prompter without adding useful meaning to the request. Examples might include repeated combinations of workplace, location, dates, habits, or other indirect identifiers. This requires conversation-level tracking and careful privacy design, so it is not part of the current local demo.
Log in or sign up for Devpost to join the conversation.