Vectara

May 09 , 2024 by Ofer Mendelevitch

Introduction

In April 2024 we held a virtual, worldwide hackathon, in partnership with lablab.ai, Together.AI, Unstructured.IO, LlamaIndex, and Tonic.AI. 

The “advanced RAG” hackathon attracted 1205 super-talented hackers, and we had 206 teams formed during the hackathon with 47 final submissions.

The results were incredible, and we’ve witnessed creative power unleashed exploring how advanced RAG techniques can be applied to build Generative AI applications, utilizing Vectara as the core platform with additional capabilities from Together AI, LlamaIndex, Unstructured.IO and Tonic.

I’m excited to share in this blog post more details about the winners of the hackathon and the amazing projects they have created.

What is Advanced RAG?

RAG as a methodology continues to be a common and a highly useful approach for building Generative AI applications such as chatbots, question-answering, or research applications.

In 2023, there were only a few tools available for building such applications, and for the most part, one had to set up each component in the stack directly. 

Since then, things have improved in two main ways. First, RAG-as-a-service platforms like Vectara provide this full stack behind an API, dramatically simplifying the GenAI development process, and making RAG easier to deploy at enterprise scale. 

In addition, we now understand a lot better, as a community of LLM developers, how to build RAG applications: 

• Ingesting the data for your RAG pipeline may require connecting to various data sources – we partnered with Unstructured.io, Airbyte, and Datavolo, as well as added exciting capabilities (such as a shiny new slack crawler) to the open-source vectara-ingest project to provide multiple options to developers for low-code data ingestion.
• We launched an integrated Chat functionality in Vectara that automatically handles chat history and ensures it is fully integrated into your Chatbot application. 
• We introduced HHEM, the open-source Hallucination evaluation model, available on HuggingFace.
• We created multiple open-source tools that help create modern user interfaces for GenAI applications such as Create-UI, vectara-answer, react-chatbot, react-search, and JS-widgets.

Combining these new capabilities with additional capabilities from Together.AI, LlamaIndex, Unstructured.IO, and Tonic, provided a strong set of technologies for hackers to use in this hackathon.

Sidebar: RAG and Long Sequence LLMs

A week or two before this hackathon, we had an opportunity to host a panel about a topic that intrigues many: does the fact that LLMs now support long sequence lengths mean RAG is no longer important or will eventually die?

Well, the TL;DR is: absolutely not. In fact – RAG will only get better!

We had a great panel of experts from Vectara, Together.AI, and LlamaIndex discuss this topic in-depth, and I invite you to watch the video if you’re interested in learning more.

Hackathon Winners

Without further ado, I’d like to share the details of our winners for this hackathon.

The first place was taken by team Puppeteers with their note-taking app MindPal.  Combining Vectara, streamlit, and Together.AI, Yasmine and Melek built a great application that showcases how RAG can help you organize your notes and more importantly access the knowledge in those notes via a chat-bot interface powered by RAG.

In second place we have Garden Rules with the Gardening chatbot. Addressing the complicated rules and regulations around gardening in Germany, and empowering the gardening community to thrive through access to a specialized chatbot, powered by Vectara, Together.AI, and Tonic. Congratulations Richard!

The third place prize goes to team Blazzbyte who had built the Legallit application, an intelligent legal assistant that can help with contract review and compliance management, built with Vectara, Together.AI, LlamaIndex, and Unstructured. Nice Job!

In addition to these overall winners, we had another 5 special prizes:

• Best use of Vectara Chat: project Wisdom Hoard. 
• Best use of Together AI: project Super Human AI.
• Best use of LlamaIndex: project Issue Co-pilot.
• Best use of Unstructured.io: project Snowflake.
• Best use of Tonic.AI: project hirechat.

A big congratulations to all the winners!

Conclusions

Hackathons are a great way not only to learn about technologies like LLMs and platforms like Vectara, but also highlight innovative solutions, tickle our creative minds to find solutions to important problems, and allow teams to come together and build great things. And sometimes these projects continue to evolve in real businesses. 

We feel fortunate that we had the opportunity to put this hackathon together with our partners and see all the great projects that were built by an amazing group of developers.

If you didn’t get a chance to participate in the hackathon we encourage you to learn more about Vectara and look at the great work shared by the winning teams as well as all the other teams, on the hackathon page.

Here are some additional resources:

• Sign up for a free Vectara account.
• Read our documentation to learn more about using Vectara.
• Learn how developers in financial services, healthcare, legal, and Education are using Vectara to build powerful RAG applications.
• We’ve created lots of sample code on our github repository for you to use, and some sample applications here and here.

And please join our developer community on our Discord server or in our discussion forums.