RAG backend using fastAPI with no gui (http://127.0.0.1:8000/docs#/default/)

Run: pip install -r requirements.txt

Setting Up Vector Storage

PGVecto.rs

1. Run the docker pgvecto-rs image, download if needed below

docker run \
  --name pgvecto-rs-demo \
  -e POSTGRES_PASSWORD=mysecretpassword \
  -p 5432:5432 \
  -d tensorchord/pgvecto-rs:pg16-v0.2.0

2. Setting environmental variables Must set VECTOR_DB_MODEL=pgvecto_rs

3. Env variables will be set as below for default credentials, modify if needed (no action needed for running docker image above) DB_HOST = localhost, DB_USER = postgres, DB_PASS = mysecretpassword, DB_NAME = postgres, DB_PORT = 5432,

qdrant

Set Env variables: VECTOR_DB_MODEL=Qdrant SERVER_URL="http://localhost:6333/"

1. Start Qdrant docker image in port 6333
2. Check out http://localhost:6333/dashboard to make sure DB working

Using RAG API

Starting Server

1. Make sure vector storage is running (see Setting Up Vector Storage)
2. Set env varOPENAI_API_KEY to your unique api key
3. Run uvicorn main:app --reload to start fast api server
4. Rag api access at http://127.0.0.1:8000/docs#/default/

RAG on text documents (.pdf, .txt, .docx)

1. Use POST /document endpoint
2. For file upload a .pdf, .txt, or .docx file
3. Strongly recommend setting file_id for filter key for vector search
4. Optional metaJSON meta data example: {"date":"1/1/2024",...}
5. If 200 response, above document now embeded in vector storaged, able to be retrieved
6. Go to GET /rag/ endpoint
7. Enter query will match vectors on meaning similarity
8. Enter k (defaults to 5) number of document chunks returned
9. Optionally filterJson JSON, keys to filter for
10. file_ids not needed but encouraged, list of file ids tied to documents.
11. Should return text of document chunks that are most similar to query

RAG on webpages and youtube links

1. Use POST /link endpoint
2. For link paste webpage or youtube URL
3. Strongly recommend setting file_id for filter key for vector search
4. Optional metaJSON meta data example: {"date":"1/1/2024",...}
5. If 200 response, above text of webpage or youtube transcript now embeded in vector storaged, able to be retrieved
6. Go to GET /rag/ endpoint
7. Enter query will match vectors on meaning similarity
8. Enter k (defaults to 5) number of document chunks returned
9. Optionally filterJson JSON, keys to filter for
10. file_ids not needed but encouraged, list of file ids tied to documents.
11. Should return text of document chunks that are most similar to query

Creating and Executing SQL on Excel and CSV files

1. Use POST /table endpoint
2. For file upload an excel (.xslx) or .csv file
3. set table_name to meaningful and relevant table name (used for sql generation)
4. If 200 response, table is stored locally
5. Go to GET /qa
6. Enter question
7. AI will look at table names, column names to generate and execute a SQL statement to answer given question