📃 Chat Your PDF!

(This project is a variant from https://github.com/RedisVentures/gcp-redis-llm-stack modified to use Azure OpenAI)

This example Generative AI application demonstrates how to build a simple chatbot powered by Redis, LangChain, and Azure OpenAI. It contains the following elements:

• Redis used as vector database with vector similarity search using Redis Enterprise, Redis Enterprise Cloud, or simple Redis Stack as Docker container.
• It is combined with the Python RedisVL client for Retrieval-Augmented Generation (RAG), LLM Semantic Caching, and chat history persistence

• ☁️ Azure OpenAI models for embedding creation and chat completion
• ⚙️ LangChain for app orchestration, agent construction, and tools
• 🖥️ Streamlit for the front end and conversational interface

The resulting application is an interactive chatbot built using vector similarity, GenAI text embeddings and retrieval augmented generation from PDF such as a standard brochure for cars that also keeps semantic caching to optimize LLM costs and improve user experience and context relevance with chat history.

Key use cases with Redis vector database

Redis is well-versed to power chatbots thanks to its flexible data models, query engine, and high performance. This enables users to leverage Redis for a variety of gen AI needs:

• RAG -- ensures that relevant context is retrieved from Redis as a Vector Database, given a users question
• Semantic Caching -- ensures that duplicate requests for identical or very similar information are not exhuastive. Ex:

  streamlit    | Full Response Time (secs) 1.6435627937316895
  streamlit    | Cache Response Time (secs) 0.11130380630493164

• Chat History -- ensures distributed & low latency access to conversation history in Redis Lists

Setting up the app environment

Setup Azure OpenAI models and deployments

You need to setup a deployment for your Azure OpenAI which will give you an OpenAI endpoint and API key, and then deploy models such as for embeddings and for text generation.

Update LLM environment configuration

The project comes with a template .env.template file with the following values. Make a copy of this as .env file in the same folder. Update the values below accordingly.

Please note how the variable maps to the deployment your created in Azure OpenAI.

OPENAI_API_BASE = "https://ava-openai.openai.azure.com"
OPENAI_API_KEY = "1234567890abcdefghijklmnopqrstuvwxyz"
OPENAI_AZURE_EMBEDDING_DEPLOYMENT = "ava-model-embedding"
OPENAI_AZURE_LLM_DEPLOYMENT = "ava-model-text"

OPENAI_API_TYPE = "azure"
OPENAI_API_VERSION = "2023-05-15"
OPENAI_LOG = "debug"

Setup your Redis database

You can use

• Redis Stack as a standalone database running locally or as provided in the docker-compose.yml with Redis Insight on port 8001
• Redis Enterprise as self-managed clustered on machines, VM, or Kubernetes. Please refer to doc or reach out to us - this is the most feature-complete, enterprise-grade Redis.
• Redis Enterprise Cloud for which you can get a free trial. Make sure to create a database with Redis Search (or Redis Stack) profile. It's the same as Redis Enterprise but fully managed on your cloud of choice!
• Azure Cache for Redis Enterprise. Make sure to create a database with Redis Search.

Update the REDIS_URL based on your deployment

# if using the default "docker compose" with Redis Stack running in the "redis" container
REDIS_URL="redis://redis:6379"
# if running with Azure Cache Redis Enterprise
REDIS_URL="redis://:[email protected]:6379"
# if running code locally and running a standalone Redis Stack
REDIS_URL="redis://localhost:6379"
# if running code in docker container but have Redis running on your host machine
REDIS_URL="redis://host.docker.internal:6379"
# if connecting to Redis with TLS
REDIS_URL="rediss://..."

Example setup

Redis Enterprise Cloud: Example of a fully persistent Redis database with Redis 7 and Redis Stack profile including Redis Search for vector similarity search.

Azure Cache for Redis Enterprise: Example of a 10 nodes, 500GB capacity with additional HA (total 1TB database) using E100 x 10 and Redis Enterprise clustering policy with Redis Search.

Run

To run the app, follow these steps:

1. Clone this repository to your local machine.
2. Set up your Azure OpenAI.
3. Set up your Redis vector database.
4. Copy the .env.template to .env and configure the values as outlined above.
5. Run the app with Docker compose: docker-compose up. You may comment out the redis container if you not using a local docker setup for your Redis database.
6. Hit your brower at http://localhost:8080/.

Using Redis Insight

Using Redis Insight you can observe the data model in Redis:

• content and vector embedding from tokenization of the PDFs for retrieval augmented generation with the LLM using Redis vector similarity search
• semantic caching also using vector embedding from the user questions and improving overall performacne and decreasing LLM costs.
• session storage keeping the human & AI chatbot interaction to further enhance auditability and compliance, improve models and prompt engineering as needed, or act as LLM memory.

Running the app locally without Docker

It can be useful to also run the app locally if you modify the code and want to quickly test it:

# install or upgrade pipenv
python3 -m pip install -U pipenv

# install the dependencies
# pipenv --rm
pip install --user pipenv
pipenv install -r requirements.txt

# run the Streamlit app locally
pipenv shell
streamlit run main.py --server.port 8080 --server.enableXsrfProtection false