Skip to content

JoeCat1207/ProjectCaravel

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

9 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Named after the Civ-5 Caravel... a swift and agile ship

Project Caravel is my entry into a 3 Week Hackathon with the theme of 'Building Connections'.

TO NOTE THAT THIS CODE IS SOMEWHAT BUGGY AND INEFFICIENT.

Claude API also doesn't run perfectly smooth, and this code provides very little token and time wiggle room so sometimes it does not respond. Feel free to play around, it is not anywhere near perfect but has been fun to use.

When running this project locally, the web UI recieves user input then follows this pipeline:

User Prompt -> Claude 3.7 Sonnett w/ parsing system prompt that splits the prompt into 4 categories (General knowledge, mathematics, coding and reading) -> Claude 3.7 creates prompts for each category -> Prompts are logged -> Prompts are sent to their respective LLMs (In this case, Claude 3.5 Haiku, Gemini 2.0 Flash, and Claude 3.7 Sonnet) -> Responses are combined into a singular response by Claude 3.7 w/ system prompt -> Final combined response is displayed.

Originally, this project was supposed to live in a Docker container. However, I'm not the brightest person and can't make it work on my machine.

Instead, you will need to cd into the backend and frontend folders individually in a Powershell (w/ admin privileges) instance.

For the backend:

cd backend_path pip install -r requirements.txt python main.py

For the frontend:

cd frontend_path npm install npm start

A locally hosted webserver should start at localhost:3000

You will also need to create an api_keys.env file with your api keys as such:

GEMINI_API_KEY= your key ANTHROPIC_API_KEY= your key

#Optional controls for whatever you need #directories for logging #PARSED_PROMPTS_DIR=./logs/parsed_prompts #GENERATED_PROMPTS_DIR=./logs/generated_prompts

#Rate limiting settings in case I run out of credits lol #RATE_LIMIT_REQUESTS=5 #RATE_LIMIT_WINDOW=60

About

tri-model AI aggregator

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published