PenetrateAI

Inspiration

We were inspired by Devin AI. We also wanted to build something that would help out in the field of cybersec.

What it does

PenetrateAI helps the user keep their website safe by pen testing their website to find vulnerabilities.

How we built it

We used a stack of Next.js, Tailwind and React for the Frontend. We used flask, docker and websockets for the backend. We used the websocket server to communicate between the docker image and the client. The core of our app was the GPT 3.5 LLM, we used Tavily API as a tool for our LLM so that it could access the web, we also connected the LLM to a CLI tool so that it could send commands. We also used some Lang Chain tools to further improve our LLM. Finally, we used a HuggingFace BERT model for queries.

Challenges we ran into

Some of the challenges we ran into were was that it was very difficult to configure docker to work for our purposes. It was also difficult to communicate between the proxy server and the docker image, as we had many errors while attempting the connection. It was very difficult to establish a connection because the two devices kept connecting to different access ports.

Accomplishments that we're proud of

Learning about the ins and outs of everything in Lang-Chain. Used GPT 3.5 to generate information.

What we learned

We learned a lot about Lang-Chaing as it was one of the most core technologies in our application. We also learned a lot about docker and docker images. We also learned a bit about networking because

What's next for PenetrateAI

We are planning on using a better LLM such as GPT-4 to build our app around. We are also planning on getting our own access port so that we don't need to use a mobile hotspot to be able to communicate between two devices.

Built With

• docker
• flask
• huggingface
• langchain
• next.js
• openai
• react
• shadcdn
• tailwind
• tavily
• websockets