Qlert aims to revolutionize AI usage by ensuring responsible and ethical adoption through the integration of automatic rule generation and real-time risk assessment mechanisms. By providing administrators with the tools to monitor user interactions and mitigate potential risks, Qlert promotes trust, transparency, and compliance in AI usage within organizations.
Qlert utilizes a Summarizer Language Model (LLM) to automatically generate rules from uploaded PDFs, with provisions for manual rule addition by administrators. These rules are then stored in a Qdrant Vector DB as embeddings, ensuring efficient storage and retrieval. Azure database is then used to store all queries made to LLMs and the the corresponding risk levels of these queries. Below are some screenshots showcasing different features of Qlert:
User queries trigger a similarity search on the vector DB to check for risky prompts, while output from LLM models also undergoes a similarity search. Risky prompts exceeding a threshold trigger alerts, with administrators notified via the admin dashboard. Simultaneously, users receive error responses to mitigate potential risks.
- Real-time Risk Assessment:
Qlert provides administrators with a comprehensive admin dashboard where they can view users querying risky content, monitor real-time graphical views of risky queries, and assess prompt injection attempts. Admins also have the authority to block user access if necessary.
- Admin Dashboard:
- Ensuring Responsible AI Usage
- Mitigating Potential Risks
- Enhancing Administrative Oversight
The final result of Qlert is a robust and scalable platform that promotes responsible and ethical AI adoption within organizations. With its automatic rule generation, real-time risk assessment, and comprehensive admin controls, Qlert enhances trust, transparency, and compliance in AI usage.
Qlert_Demo.mp4
Follow these steps to install Qlert on your local machine:
-
Clone the Qlert repository to your local machine:
git clone https://github.com/Sreehari78/Qlert.git
-
Navigate to the project directory:
cd Qlert -
Install the required dependencies. It is recommended to set up a virtual environment before installing dependencies:
python -m venv venv source venv/bin/activate # On Windows, use 'venv\Scripts\activate' pip install -r requirement.txt
-
Run the development server:
python server.py
-
Run client server:
npm i npm run dev
-
Open your web browser and navigate to http://127.0.0.1:3000/ to access Qlert.
- Upload PDFs to automatically generate rules.
- Administrators can manually add and delete rules.
- User queries trigger a similarity search for risky prompts.
- Alerts are triggered for risky prompts exceeding a threshold.
- Monitor users querying risky content.
- Real-time graphical views of risky queries.
- Assess and control prompt injection attempts.