BIX: a conversational AI platform for cybersecurity risk management
Balbix
The challenge
Security teams were spending significant time manually querying knowledge bases and cyberattack surface management tools to understand risk and prioritize action, slowing down response times.
Our approach
We led the design and development of BIX, integrating LLMs for natural language understanding, knowledge base querying, and action recommendations. We built robust frameworks to streamline training, testing, and deployment of LLMs, using Airflow for scheduling and LoRA, gradient accumulation, and gradient checkpointing for optimization. We also pioneered using LLMs to predict tactics and techniques associated with vulnerabilities, exploring quantization and deploying on Triton Inference Server with vLLM.
Results
- Conversational AI platform improved analyst efficiency and user experience
- LLM-based prediction of attack tactics and techniques tied to vulnerabilities
- Quantized models deployed on Triton Inference Server with vLLM for low-latency inference
Have a similar challenge?
Let's discuss how we can help you achieve results like these.