Cybersecurity Chatbot with Rasa and Large Language Models
Project Overview
This project aimed to build a user-friendly cybersecurity chatbot that answers queries about security threats and best practices. Leveraging the Rasa framework for natural language understanding (NLU) and dialogue management, I integrated large language models (LLMs) like Cohere to enhance responses.
Technologies Used
- Rasa for NLU and chatbot dialogue flow
- Cohere LLM for contextualized answers
- Python for backend integration and preprocessing
Challenges
Initially, integrating LLMs like OpenAI and Google’s PaLM API was problematic due to quota limits and paywalls. Hugging Face Transformers was functional but provided unsatisfactory responses. Ultimately, Cohere proved to be the most effective, offering accurate, context-aware responses.
Screenshots
Results
The chatbot handles user queries on cybersecurity threats like phishing and malware effectively. It successfully provides contextual recommendations while gracefully managing ambiguous queries through fallback mechanisms.
Future Work
- Expanding the training dataset to cover more specialized cybersecurity topics
- Incorporating more robust fallback handling
- Deploying the chatbot for real-world testing and collecting feedback for further improvement.