---

AskMe

An intelligent FAQ assistant that accurately answers user queries by matching them with semantically similar questions from a predefined FAQ database. When a close enough match isn't found, the system seamlessly interacts with the OpenAI API to generate the most relevant response.

Table of Contents

• About
• Features
• Project Structure
• Prerequisites
• Installation
• Usage
• API Documentation
• Monitoring
• Testing
• FAQ Database
• License

About

AskMe is a web-based FAQ assistant powered by NLP models and PostgreSQL's pgvector for embedding similarity search. It provides users with precise answers by querying preloaded FAQs stored in a PostgreSQL database and can be extended using OpenAI's GPT models for more complex queries.

Features

• FAQ Querying: Get precise answers to commonly asked questions.
• Embedding Computing: Compute embeddings from the provided FAQ data for later similarity searches.
• LangChain: Utilize LangChain for NLP tasks outside of OpenAI interactions.
• OpenAI Integration: For queries that don't match any existing FAQ, the system can generate responses using OpenAI via LangChain.
• FastAPI Endpoint: Provide an API for users to submit questions and receive answers.
• Authentication: Added to endpoints using FastAPI’s dependency mechanism.
• PostgreSQL for FAQ and Embeddings: Integrated pgVector to store and query vector embeddings.
• Containerized Deployment: Full Docker support for easy deployment.

Project Structure

.
├── app
│   ├── scripts
│   │   ├── initialize_embeddings.py
│   │   └── load_faq_data.py
│   ├── services
│   │   ├── config.py
│   │   ├── embeddings.py
│   │   ├── openai_client.py
│   │   └── similarity.py
│   ├── static
│   │   ├── favicon.ico
│   │   ├── main.js
│   │   └── styles.css
│   ├── templates
│   │   ├── error.html
│   │   └── index.html
│   ├── utils
│   │   └── logger.py
│   ├── db.py
│   ├── main.py
│   ├── models.py
│   └── init.sql
├── data
│   └── faq_data.json
├── tests
├── .env_example
├── Dockerfile
├── docker-compose.yml
├── requirements.txt
└── README.md

Prerequisites

Before you begin, ensure you have the following installed on your system:

• Python 3.10+
• Docker

_{Note: You will also need an OpenAI API key to interact with OpenAI.}

Installation

To install and run AskMe app using Docker, follow these steps:

1. Clone the repository:

git clone https://github.com/lkmeta/askme.git
cd askme

2. Set Up Environment Variables

cp .env_example .env

_{Edit the .env file and add your API keys for OpenAI API, Authentication, and PostgreSQL necessary environment variables.
Additionally, you can configure the following parameters:
EMBEDDINGS_MODEL: Define the model for embeddings (e.g., text-embedding-3-small).
SIMILARITY_THRESHOLD: Set the similarity threshold for searching (default: 0.7).}

3. Build the Docker Image

docker-compose build

4. Run the app

docker-compose up

_{Note: Upon the first run, init.sql is automatically executed, initializing the database with the required schema. The load_faq_data.py and initialize_embeddings.py scripts will load the FAQs and embeddings.}

Usage

Access the Application

Open your web browser and navigate to http://localhost:8000 to access AskMe.

API Documentation

FastAPI automatically generates interactive API documentation. Once the application is running, access the documentation at:

http://localhost:8000/docs

Monitoring

To monitor the application and the services:

1. Ensure you have completed the installation steps above.

2. You can view the logs of the running Docker container to monitor the application output.

docker-compose logs -f

_{Note: The -f option follows the log output in real-time.}

Testing

To benchmark the performance of the AskMe application, you can use Locust, a load-testing tool that helps simulate concurrent users sending requests to your API.

The locustfile.py is located inside the test directory.

Steps to Run Locust Test

1. Install Locust: If you don't have Locust installed, you can install it using pip (preferably within a virtual environment such as python env):

   pip install locust

2. Navigate to the Test Directory:

   cd test

3. Run Locust: Start Locust by running:

   locust -f locustfile.py

4. Open Locust Web Interface: Open your web browser and go to http://localhost:8089.

5. Start the Test: Fill in the required fields (like the number of users) and set the Host to http://localhost:8000 to begin testing the AskMe application.

FAQ Database

The predefined set of FAQ questions and answers used to compute the embeddings is located in the file data/faq_data.json. You can update this file with additional FAQs to improve the system’s accuracy.

License

This project is licensed under the Apache-2.0 License.