chatterbox

Warning

At this time, this is a personal project and not intended for distribution.

My primary use for generative ai leveraging large language models is for scientific research and code development. While today's llm models are quite adept at solving most problems, I often would like to feed research articles and/or open source projects to the llm for additional context. Many of those research articles are likely outside the scope of the llm's training data.

Chatterbox is a collection langgraph workflows (graphs) made up of components (nodes, conditional edges, and utilities). Each workflow has an associated frontend web application built with Streamlit.

Run the app to launch stremlit chat (requires an .env file with API KEYS)

uv run python app.py --chat

Or

uv run --env-file .env -- python st_chat.py

large language models

One of the objectives of the project was to explore different large language models for different agents. Chatterbox does this by providing a function get_llm_model that returns a BaseChatModel for each llm defined in the LargeLanguageModelsEnum.

See language_models.py

requirements for using large language models

To use the language models provided by Anthropic, OpenAI, or Fireworks, requires an api key which must be stored in the .env file.

research notes to vector databases

Collecting information from the web, arxiv, and pdf documents will ususally provide the context necessary to answer a question, if you get the chunking right and have enough context in each document.

Another option is to build a collection of documents from research notes. To do this, I write my research notes in latex files (.tex) and use the ... to load and save to a Chroma database.

Chat

The chat app is the simplest application ...

To run,

uv run python app.py --chat

Research Context Summary

The objective of this graph is to summarize all relevant documents to the input research prompt.

agents (nodes)

triage

research context supervisor

pdf context retriever

web context retriever

arxiv context retriever

vdbs context retriever

research context grader

research context summarizer

discussion and future work