Updates :
XShield is now developed and deployed fully using Nvidia AI Workbench. We have leveraged the compose feature of Nvidia AI Workbench to manage our multicontainer AI application. This improvement after the valuable feedback of the judges has made our application a lot more easier to setup and deploy.
XShield is now integrated with Open source Static Application Security Testing tool Bandit to analyze and generate Static Testing report. This gives the LLM additional context and allows it to create more accurate and detailed security reports.
Now XShield provides insightful fix suggestions and also fix code snippets to solve both security and compliance issues for the client.
We also implemented a multi LLM infrastructure for xShield. Now users are able to choose an foundation model of their choice. Right now we are offering locally hosted Llama3 and GPT4 through Openai API.
We also implemented a caching system for XShield using redis to cache contexts and reports of code files. Since LLMs have limited token and context windows, we have developed an Intelligent system that only processes related and dependencies together to generate correct analysis. Since most files remain unchanged, the caching system reuses the previously generated reports and context information for these file avoiding duplicate analysis on the LLM saving compute resources.
Now xShield also has an effective system that could reduce the json formatting generated by smaller LLM Models like LLama3. The JSON fix system first uses a json fixing library followed by GPT4 to further fix json errors.
We have refactored certain sections of the code and also removed hardcoded parameters.
Log in or sign up for Devpost to join the conversation.