EstephaniaCalvoC/celeste-mvp-back

Celeste chat bot's backend

Python

celeste-mvp-back

API to interact with a Chatbot to answer questions about a markdown file or a web page using ChatGPT from OPenAI.

Status: In progress

How to run

Set environment variables in .env

# Mandatory
OPENAI_API_KEY="openai_api_key"

# For Monitoring (Optional)
LANGCHAIN_API_KEY="langchain_api_key"
LANGCHAIN_TRACING_V2='true'
LANGCHAIN_ENDPOINT='https://api.smith.langchain.com'

Set confing in .config.json

{
  "cache_path": "Directory path where will be the embeddings stored",
  "files": ["file1.md", "./my_knowledge_file2.md"],
  "pages": [
    {"url": "https://example.com", "tag": "div", "class": "my-class", "id": "my-id"},
    {"url": "https://other_example.com", "tag": "body", "class": "your-class", "id": "your-id"}
  ]
}

Either files or pages must be in the configuration.
url is mandatory to define pages.

Set virtual environment

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Run service
```
python app/serve.py
```

How to call

Using request

import requests

response = requests.post(
    "http://localhost:8000/markdown/invoke",
    json={'input': {"input": "Are Paper Checks processed any day?"}}
)
print(response.json()["output"]["answer"])

Using RemoteRunnable

from langserve import RemoteRunnable

 
def main():
    """Client to interact with the served chain as if it were running client-side"""
    remote_chain = RemoteRunnable("http://localhost:8000/markdown")
    response = remote_chain.invoke({"input": "What is a billpay?"})
    print(response)
    
if __name__ == "__main__":
    main()

Using the Navigator

Open http://localhost:8000/rag/playground in the navigator.

How to contribute

TODOs

MVP-001: Load documents from page
MVP-002: Test basic base of knowledge
MVP-003: Expand the base of knowledge
MVP-004: Enable a chat mode
MVP-005: Use an open source model

Tests cases

Markdown base of knowledge - basic
HTML base of knowledge - basic
Expanded base of knowledge