Smarty Gemini

• 

  using arxiv and google-scholar search tools

• 

  using wikipedia and pub_med tools

Inspiration

The dream of creating my own AI assistant akin to Iron Man's JARVIS ignited my passion for AI, specifically LLMs and custom agents. This vision drove me to participate in a hackathon hosted by Google, utilizing the Gemini API, and set me on a path of exploring AI technologies.

What it does

Smarty Gemini serves as an intelligent AI assistant who can use many tools to process and retrieve information. It will use DuckDuckGo, Wikipedia, Arxiv, Google Scholar, and PubMed Basic Search to read and interpret the PDF files and answer questions from the findings of the PDF files in a unique way. Their goal is for the information flowing in to appear quite seamless and intuitively helpful for the user.

How we built it

The project was based on LangChain and LLM API calls, but more than that, it is dealing with the integration of Gemini, along with a useful set of tools. They extended the features of Gemini by building a custom tool to read and extract information out of PDF files. We utilized the Streamlit library to create a very simplified and easy interaction platform with the users over the AI features.

Challenges we ran into

The system prompt capable of effectively guiding and passing LLM responses was for us the most difficult one to be formulated. This, of course, demanded a highly precise and clear formulation, able to guide Gemini properly and coherently in every taken up task.

Accomplishments that we're proud of

Being able to extend Gemini to read and answer questions from a PDF file is counted as part of some of the best things in this project. Two of the most important milestones were deploying on the Streamlit Community Cloud server and publishing our code on GitHub. These successes have been nothing but the realization of a dream; one's own as well as contributing to the realization of a dream for the rest of the members.

What we learned

This project has given us further understanding regarding LLMs, API integration, and how we can develop a custom tool to extend AI agent functionality. It also taught me, through the system prompt, how critical communication is and the challenge to provide an intuitive user interfac

What's next for Smarty Gemini

In the future, we hope to continue refining the functionalities of Smarty Gemini and develop more integrations and tools that will improve its utility. Therefore, Smarty Gemini will be much more flexible with the advance of technology in the way it was going to be used to base user feedback on continuous improvement.

Built With

• faiss
• langchain
• python
• streamlit