Inspiration
As a Certified Ethical Hacker (CEH) and AI enthusiast, I’ve always been fascinated by the intersection of cybersecurity and artificial intelligence. During my internship, I worked on a generative AI-based cybersecurity Q&A system, which planted the seed for SecureGPT. I wanted to build a tool that could bridge the gap between complex security data and accessible, actionable insights — especially for non-expert users and small IT teams.
What it does
SecureGPT is a conversational cybersecurity assistant that:
Analyzes log files (e.g., syslog, auth.log, web server logs). Detects anomalies and suspicious behavior using embedded threat intelligence. Leverages Retrieval-Augmented Generation (RAG) to answer user queries like: "Was this IP part of a brute-force attack?" "Is this log line suspicious?" Recommends defensive actions or patches. Acts as an educational tool by explaining security concepts in plain English.
How we built it
FastAPI + LangChain form the backend core, offering a flexible framework for managing prompt flows, tool integrations, and real-time interactions. A React-based frontend powers the user interface, enabling smooth, real-time chat with the AI agent. Implemented Retrieval-Augmented Generation (RAG) using a vector database Pinecone to ground responses in contextual cybersecurity documentation and threat intelligence.
Challenges we ran into
Prompt grounding: Making GPT reliably identify malicious patterns without generating false positives. Context limitation: Handling long log files required intelligent chunking and summarization strategies. Latency: Speed-optimized the retrieval layer to ensure the app remained responsive.
Accomplishments that we're proud of
What we learned
Gained hands-on experience integrating RAG pipelines for a domain-specific task (cybersecurity). Improved my skills in prompt design, model alignment, and handling hallucination risks. Learned how to preprocess and vectorize large log files efficiently.
Log in or sign up for Devpost to join the conversation.