issacchan26 / talktopdf

RAG serverless function with AWS Lambda, OpenAI API and Pinecone vector database

AWS Lambda RAG serverless function with Pinecone vector database

This repo provides RAG serverless function connected with Pinecone vector database. The user can upload PDF files to Pinecone and input user prompts to this function to extract useful information.

Build your own Pinecone vector database

Please run this notebook to create the Pinecone vector database with your own PDF. It is modified from this notebook by adding dependencies.

Docker

The AWS Lambda function is built with Docker. This repo provides Dockerfile and requirements.txt to build the Docker image.
First, please make sure that you have set up your IAM user with AWS CLI, please refer to https://docs.aws.amazon.com/cli/latest/userguide/cli-authentication-user.html for instructions.
If this is the first time you start with IAM, please make sure that you add AmazonEC2ContainerRegistryFullAccess and AmazonECS_FullAccess to your IAM user.
And set up your Amazon ECR with the reference of https://docs.aws.amazon.com/AmazonECR/latest/userguide/getting-started-cli.html for general instructions. Then, you are ready to build and push your docker image to ECR:

git clone https://github.com/issacchan26/TalkToPDF.git
cd TalkToPDF

Authenticate to your default registry

aws ecr get-login-password --region your_region | docker login --username AWS --password-stdin your_aws_account_id.dkr.ecr.region.amazonaws.com

Create a repository named 'lambda-docker-deploy' with your region

aws ecr create-repository --repository-name lambda-docker-deploy --region your_region

Build your docker image

docker build -t lambda-docker-deploy .

Tag and push your docker image to ECR

docker tag lambda-docker-deploy:latest your_aws_account_id.dkr.ecr.region.amazonaws.com/lambda-docker-deploy:latest

docker push your_aws_account_id.dkr.ecr.region.amazonaws.com/lambda-docker-deploy:latest

Finally, you can deploy your AWS Lambda function by selecting your container image in ECR.

Create API endpoint

After deploying your AWS Lambda function, you can create the API endpoint by adding trigger of API Gateway, creating new HTTP API and configure Security to 'open'.

Usage with curl

curl [options...] [url]  
-X, request command, this function supports POST request  
-H, content type, which is "Application/JSON"  
-d, data, here we enter our userprompt as data  
url, API endpoints  

Example:

curl -X POST -H 'content-type: application/json' -d '{ "userprompt": "Who published Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks?" }' https://your_api_endpoints_of_talktopdf/ 

Enter your question by replacing the value of userprompt in option -d above, e.g.

curl -X POST -H 'content-type: application/json' -d '{ "userprompt": "What is RAG?" }' https://your_api_endpoints_of_talktopdf/

Example userprompts and responses

The example used here is the first page of:
Lewis, Patrick, et al. "Retrieval-augmented generation for knowledge-intensive nlp tasks." Advances in Neural Information Processing Systems 33 (2020): 9459-9474.

Start with a simple question

Q: What is RAG?
A: RAG stands for retrieval-augmented generation.

Modify the keywords of the prompt to get desired answer

Q: What is RAG in details?
A: RAG stands for retrieval-augmented generation. It is a model that combines pre-trained parametric and non-parametric memory for language generation. In this model, the parametric memory is a pre-trained seq2seq model, and the non-parametric memory is a dense vector index of Wikipedia, accessed with a pre-trained neural retriever. There are two formulations of RAG: one that conditions on the same retrieved passages across the whole generated sequence, and another that can use different passages per token. RAG models have been shown to generate more specific, diverse, and factual language compared to state-of-the-art parametric-only seq2seq models.

The model can only answer the question in the PDF

Q: What is the color of an apple?
A: Based on the given context, it is not possible to determine the color of an apple.

The model can provide accurate and useful information of the PDF

Q: Who published Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks?
A: The authors of "Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks" are Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen-tau Yih, Tim Rocktäschel, Sebastian Riedel, and Douwe Kiela.