New
Introducing Compute Orchestration
October 9, 2023

Meet the Clarifai Champs of the Streamlit LLM Hackathon

Table of Contents:

Winners Streamlit Blog Header (1)

In September, Clarifai was excited to be part of an LLM Hackathon hosted by Streamlit. The Hackathon was intended to nurture creativity in the development of AI-first products using powerful LLMs.

Building with LLMs and other models from Clarifai community

The challenge required participants to create a Streamlit app by utilizing technology from Clarifai and four other partners: LangChain, AssemblyAI, Weaviate, and LlamaIndex.

A total of 36 impressive projects were built using Clarifai and submitted for judging. With so many great entries, selecting two Clarifai winners proved challenging! Here are the projects that stood out the most!

Winner: GPT Research Buddy

GPT Research Buddy, one of the winners, is a Q&A and summarization bot leveraging the Nougat Transformer to read research papers from sources like Arxiv and the ACL Anthology. It uses LLamaIndex to create a vector-based index for Q&A discussions and also leverages GPT 3.5 Turbo, Open AI's Text Embedding, and Text Moderation models from the Clarifai Community, along with the Nougat Transformer.

Here's how GPT Research Buddy works:

  1. With the user-provided URL of the arXiv or ACL Anthology research paper, it uses Nougat Transformer to read the paper contents and get the markdown representation of the paper.
  2. Next, it uses LLamaIndex to create a vector-based index for Q&A discussions to generate key insights from the paper, parallelly spinning up a custom chatbot engine.
  3. Now, you can ask the chatbot questions specific to the research paper. Before answering, the query will be passed through a moderation engine, which decides whether the query is toxic, hateful, or biased, using Clarifai's text moderation model.

Screenshot 2023-10-05 at 4.21.08 PM

Winner: Persona Simulator

Persona Simulator features five different personas, each with their own personalities, recent actions, goals, and a set of available tools to determine their actions. These personas can create artwork, compose blog posts, and share social media updates. The project utilizes the Stable Diffusion XL 1.0 and GPT-4 models from the Clarifai Community.

Here's how Persona Simulator works:

The app contains several distinct personas, each represented by a paragraph of text, a list of interests, a physical description, and a writing style description. It also has access to a variety of tools for internal reflection and external expression. Additionally, a workflow has been established in which each persona is guided independently to select a task based on the tools they can access. Subsequently, they plan how to execute the task and then proceed to do it.

To get started, you select a persona of your choice and give them their next thought. As it progresses, you'll be able to see their thought process as they gather research, journal about their emotional response, create art and make various types of postings.

The tools available to the five AI personas include:

  1. Read the latest blogs: Retrieves the contents of the last three blog posts for continuity.

  2. Read from journal: Retrieves the contents of the last three journal entries.

  3. Create art: Generates various forms of art, such as photography, paintings, and digital art.

  4. Write in journal: Records information in a journal, facilitating thought processing after tasks like QueryForInfo.

  5. Post on social media: Shares short messages on social media, optionally including images created using CreateArt.

  6. Write blog post: Enables writing long-format blog posts for immediate publication.

  7. Query for info: Serves as the primary interface for asking questions and acquiring information to be used in blog entries.

Screenshot 2023-10-05 at 4.23.02 PM

Congratulations to these two winners!!

Honorable mentions🏆

Check out some other awesome projects below - the Honorable Mentions. Note that they are listed in no particular order.

DocuTalk

DocuTalk is an LLM-powered chatbot that utilizes Clarifai's documentation (Clarifai docs) as its knowledge base and answers questions.  It also leverages the embedding model along with GPT-4 available in the Clarifai Community. 

Tube-to-Text-Coach

Tube-to-Text-Coach is a workout companion that transforms follow-along exercise videos into fitness routines.  Input your favorite workout video and it creates a workout routine leveraging GPT-4 available in the Clarifai Community.

Design Aid

Design Aid is a design assistant that helps guide the design process through questions, problem statement generation, product mock-ups, a sample user journey, interviews from virtual users, and report generation. This utilizes Open AI embeddings, GPT-4, content moderation, and Stable Diffusion XL models all available in the Clarifai Community. 

Final Words

Thanks to Streamlit for partnering and hosting this hackathon and a special thanks to the other partners LangChain, AssemblyAI, Weaviate, and LlamaIndex. Also congratulations to the hundreds of hackers that came together to build, learn, and have fun. We love seeing your great ideas, energy, and collaboration. Once again we are impressed with the number and quality of the projects built with Clarifai.

Want to stay in the loop for the next Clarifai Hackathon? Follow us on Twitter X and join the Discord Community to stay up to date!