Retrieval Augmented Generation IRIS, Intersystems

In this project we decided to use a vector database to build a chat that helps people navigate through huge amounts of information very fast.

Comment

Inspiration

Looking at all the challenges, this one specially caught our eye because it was making use of cutting edge technology such as ChatGPT and vector space databases. This meant that getting to work on this project would give us some insight about technologies that in upcoming future will have a tremendous impact on our life.

What It Does

Our project, IRIS (Intersystems Retrieval Augmented Generation), uses RAG principles, Python, and JavaScript to deliver precise and contextually relevant information from a given data Information set. It then uses the OpenAI API to answer the user's questions about the information that it has absorved. This means that it can allow a fully trained generative model to further it's understanding of the world without having to retrain the model with more data. This also allows to apply this cutting edge technology to many more useful situations in the real world that it would be without RAG.

How We Built It

The webApp has been built with a solid JavaScript, HTML frontend that swiftly communicates through fastAPI with a Python Backend. The frontend is in charge of getting all the information provided by the user improve the AI and the backend is in charge of managing both the vector space database and the OpenAI generative model so they can work in harmony to output the best output possible.

Accomplishments That We're Proud Of

Successfully integrated a multi-step data retrieval system that improves upon existing methods by offering highly relevant and concise information. Developed a scalable architecture that can be adapted and expanded to include additional data sources. Created a user-friendly platform with real-world usability for knowledge workers, students, and researchers. Managed to build a reusable modular code that could be imported into a library.

What We Learned

As technology concerns, during this project we mostly learned about generative AI and vector space databases. Moreover, we furthered our understanding of web development and backend. Coming right after that, we improved our knowledge of API's and (web/PDF)Scrapping. Finally, we also learned a lot about servers, Docker, Python and JavaScript. Not only did our technical skills get better but also, during all the hackathon our teamwork skill got considerably better over time.

What's Next for Retrieval Augmented Generation IRIS

To further de usability of the webapp, the future plans of the project are to become a google Chrome extension aimed at websearching. The idea is to make it become a small chat on the right part of your screen so that when you enter a new webpage, instead of navigating through the information alone the small chat can be used to query the data of the page without the trouble of searching for it's physical location.

During the last day we have already been working on making it happen and launching all the program through an online server, however, the IRIS library has been giving us some problems in the server.

Built With

Share this project:

Updates

posted an update —

Justification comments about having git commits a little after 9:00: The first one that is made after 9 is: "Update Chat.svelte". All the functionality demonstrated had already been committed before 9:00. The backend and the frontend worked separately by themselves. All the commit after 9:00 are only frontend changes and are in its entirety to solve a CORS issue we ran issue while testing it preparing for the demo.

Log in or sign up for Devpost to join the conversation.