aap-rag-content

Overview

This repository was created based on the codes found in the openshift/lighspeed-rag-content for building a RAG (Retrieval-Augmented Generation) vector database, which is used with the Ansible Automation Platform (AAP) chatbot from the documentation sources stored in the ansible/aap-docs repository.

Prerequisites

• PDM
• Asciidoctor

Setup

make install-tools
make install-deps
make install-deps-test

Build aap-rag-content image manually

Currently, aap-rag-content images are built with Gitlab.cee aap-rag-content repository, which references this repository as a git submodule. However, If you need to build an image manually from this repository, use the following steps.

Extract Markdown files from Mimir archive

1. Obtain the access to the Mimir repository and clone the repository.

2. Create the ./mimir folder in the project root.

3. Copy mimir-extract-latest.tgz.enc file to ./mimir

4. Run ./scripts/mimir-parser.py, which will extract markdown files in ./aap-product-docs-plaintext folder.

    ./scripts/mimir-parser.py
   

How to include Knowlege Base articles

If you want to include Knowledge Base articles, which are stored under the red_hat_content/solutions folder in the Mimir archive, run the mimir-parser.py with the --add-kb-articles option, i.e.,

./scripts/mimir-parser.py --add-kb-articles

The script extracts only the articles whose [products] metadata contains Red Hat Ansible Automation Platform. The metadata is defined in the beginning of each Knowledge Base markdown file.

Build image

make build-image-aap

Push image to quay.io

podman login quay.io
podman push aap-rag-content quay.io/ansible/aap-rag-content

Experiment Postgres (PGVector) Vector Store

By default, Faiss Vector Store is used for saving embeddings and the result is included in container images. You can also use Postgresql database as the vector store with its PGVector extension.

Start Postgres with PGVector extension

make start-postgres-debug

The data directory of Postgres is created under ./postgresql/data.

Generate embeddings

make generate-embeddings-postgres

The result is saved in the data_aap_product_docs_2_5 table.

$ podman exec -it pgvector bash
root@7894ab5c94e2:/# psql -U postgres
psql (16.4 (Debian 16.4-1.pgdg120+2))
Type "help" for help.

postgres=# \dt
                   List of relations
 Schema |           Name            | Type  |  Owner
--------+---------------------------+-------+----------
 public | data_aap_product_docs_2_5 | table | postgres
(1 row)

postgres=#