Project Structure

• main.py: This is the main FastAPI application file where we have registered all our endpoints.
• dependencies.py: Contains the code for loading environment variables, setting up the MongoDB connection and initializing the llms.
• prompt_templates.py: Stores all the template string for the prompts.
• prompt_examples.py: Loads all the examples for each prompt from the database.
• prompts.py: Creates all the Prompts(LangChain Format) by importing prompt templates from prompt_templates.py and examples from prompt_examples.py
• clf_api.py: This file has the enpoint for classifying the user complaint.
• rl_apis.py: This file contains the endpoints for updating the examples for the prompt.
• translate.py: This file provides the translation from english to spanish and korean.

Latest deployment instructions

Commands after SSHing into remote Azure server:

1. cd ssd/sensfix-demo/
2. git pull origin main
3. deactivate
4. rm -rf venv
5. python3.8 -m venv venv
6. source venv/bin/activate
7. pip install -r requirements.txt
8. Inside tmux session:

   8.1 Ctrl + C
   8.2 deactivate
   8.3 source venv/bin/activate
   8.4 uvicorn main:app --port 9000
   8.5 Ctrl + B. Then D.
   

9. deactivate

How to Run

1. Set up your credentials by creating a .env file in the project directory with the following content:

   MONGO_DB_URI=<your mongo db uri>
   OPENAI_API_KEY=<your api key>
   DB=<database name>
   COLLECTION_CAT_L1=<collection name for category l1>
   COLLECTION_CAT_L1=<collection name for category l2>
   COLLECTION_VALID=<collection name for valid complaint(complete/not complete)>
   COLLECTION_TICKET=<collection name for complaint ticket(ticket/not ticket)>
   COLLECTION_RESP=<collection name for complaint reponse>
   TRANSLATE_KEY=<rapid api key for google translate by google cloud>
   

   Translate API link: https://rapidapi.com/googlecloud/api/google-translate1

2. Clone this repository to your local machine:

   git clone https://github.com/T3xtifyai/sensifix-demo.git

   and move the .env file into this folder.

3. Navigate to the project directory:

   cd sensifix-demo

4. There are two ways you can run the app.

   a. Use Docker (recommended)

   first build the docker image

   docker build -t sensfix_image ./

   second run the docker image

   docker run -d --name sensfix_container -p 80:80 sensfix_image

   left part in port 80:80 is host port and the right part is the container port.

   b. Virtual environment

   python -m venv venv
   source venv/bin/activate  # On Windows, use: venv\Scripts\activate

   Install the project dependencies:

   pip install -r requirements.txt

   uvicorn main:app

5. Access the API in your web browser or use a tool like curl or httpie to make HTTP requests.

API Endpoints

1. /ticket_qualification: Categorize user complaint and generate the appropriate response
   inputs: user_input: user complaint, language, memory: previous context.
   output: ticket_status, category, and response

2. /complaint_is_complete: endpoint to update the valid complaint type(complete/incomplete)
   inputs: text: complaint text, correct_class: could be either ("complete", "incomplete").

3. /complaint_l1_type: endpoint to update the correct level 1 category of the complaint
   inputs: text: complaint text, correct_class: could be either ("electrical_it", "cleaning_janitorial", "building_infrastructure", "security").

4. /complaint_l2_type: endpoint to update the correct level 2 category of the complaint
   inputs: text: complaint text, correct_class: could be either ("electrical", "it", "cleaning", "janitorial", "building", "infrastructure", "security").

5. /complaint_response: endpoint to update the correct response for complaint with the correct category
   inputs: complaint_text: complaint text, complaint_response, complaint_class: could be either ("electrical", "it", "cleaning", "janitorial", "building", "infrastructure", "security").

6. /ticket_clf: endpoint to update the correct ticket type of the complaint
   inputs: text: complaint text, correct_class: could be either ("ticket", "not_ticket").

HTTP Status Codes used

• 200: Success
• 430: Incomplete complaint
• 431: Invalid Level 1 classification
• 432: Invalid Level 2 classification
• 504: Project API Gateway Timeout, the request is not completed within time.