DLLM Checkmate

• 

  in-person @ georgetown university

• 

  key description for DLLM Checkmate

Our project, DLLM Checkmate, was inspired by the critical security gap in modern AI development, where traditional tools fail to detect LLM-specific vulnerabilities. We built a scanner that ingests user-uploaded code files and employs a Retrieval-Augmented Generation (RAG) system enhanced by Snowflake Cortex API, comparing code against authoritative security frameworks like NIST 800 and MITRE to identify risks. The system then generates plain-language explanations, provides safe remediation fixes, and stores all findings in Snowflake for comprehensive security analytics all delivered through an intuitive, user-friendly interface. During development, we navigated significant technical hurdles, including a strategic pivot from a front-end framework to Streamlit and the complexities of integrating Snowflake Cortex API managing data latency, structuring results for real-time retrieval, and optimizing RAG performance with vectorized security benchmarks like NIST-800 and MITRE. Despite these obstacles, we successfully engineered a robust backend that trains the LLM to accurately detect vulnerabilities and propose fixes, paired with an intuitive front-end interface that clearly visualizes risks, delivers actionable remediation suggestions, and supports data-driven analytics. Through this process, we recognized the substantial value a tool like DLLM Checkmate provides to organizations and developers, offering critical preemptive security that saves time, reduces cost, and protects both infrastructure and client data. Looking ahead, we plan to expand into a comprehensive mobile and desktop application, integrate broader compliance frameworks, and develop real-time monitoring capabilities to further solidify its role as an essential guardrail in the AI development lifecycle. 1859AFD01750E58B

Built With

• mitre
• nist-800
• python
• rag
• snowflake