The goal of this demo app is to showcase the quickest and easiest way to build a usable knowledge database using Retrieval-Augmented Generation (RAG). RAG enhances the ability to answer questions by combining retrieval and generation capabilities.
Use case: seed knowledge with maintenance manuals, ask questions for troubleshooting issues. Provided is an example PDF file (t60.pdf - maintenance manual for IBM T60 laptops).
.env.example
file to .env
and fill in the required valuesBuild the Docker image:
docker build -t local-rag:latest image/
Run the Docker container:
docker run -it --env-file .env local-rag:latest python3 pdfloader.py
This will:
amazon.titan-embed-text-v2:0
model)Some example questions and generated answers can be found in the results/
directory.
Tons of them, but here are a few:
Using AWS Bedrock will incur some costs.
With the current setup (example t60.pdf file of ~200 pages mixed text and images, settings for chunking text) the whole corpus is about 200 000 tokens. Running it through the amazon.titan-embed-text-v2:0
model will cost about $0.004.
Asking questions and sending prompts to anthropic.claude-3-haiku-20240307-v1:0
results in minimal costs for input/output model tokens. For rough estimate - everything in the results/
direcory was generated for about $0.005.
Keep in mind those are one of the cheapest models, so probably a model switch will increase costs.