This project is a small end‑to‑end Retrieval-Augmented Generation (RAG) pipeline focused on InterSystems Business Services documentation.

It:

1. Chunks and embeds documentation using OpenAI embeddings (embed.py).
2. Uploads embeddings to InterSystems IRIS as a vector table (upload_embeddings.py).
3. Searches the vector table for relevant chunks (search.py).
4. Provides an interactive chat that uses OpenAI’s Responses API and your vector search as a tool (business_service_chat.py).

1. Source docs: .txt / .md files in Documentation/BusinessService.
2. Chunking & embedding (embed.py):
   • Split docs into overlapping chunks (paragraph/sentence-aware).
   • Save chunks to BusinessService.parquet.
   • Generate embeddings with text-embedding-3-small.
   • Save embedded chunks to BusinessService_embedded.parquet.
3. Vector DB (InterSystems IRIS) (upload_embeddings.py):
   • Create a vector table (AIDemo.Embeddings).
   • Upload chunk_text + embedding as VECTOR(FLOAT, 1536).
4. Search API (search.py):
   • Given a query, generate its embedding.
   • Run a similarity search using VECTOR_DOT_PRODUCT and TO_VECTOR.
   • Return top‑k matching chunks.
5. RAG Chat (business_service_chat.py):
   • CLI chat using OpenAI Responses API.
   • Exposes search_embeddings as a tool (search_business_docs).
   • Model decides when to call the search tool to answer business-service questions.

Assuming a flat structure:

• embed.py
• upload_embeddings.py
• search.py
• business_service_chat.py
• Documentation/BusinessService/
  (your .txt / .md docs go here)
• venv/dev.env
  (your environment variables – notably OPENAI_API_KEY)

• Python: 3.10+ recommended
• Install the following (adjust as needed):

pip install \
  openai \
  python-dotenv \
  pandas \
  pyarrow \
  tiktoken \
  langchain-text-splitters \
  numpy \
  iris

Note: pyarrow (or fastparquet) is needed for Parquet support; pandas will use one if available.

• You need an OpenAI API key with access to:
  • text-embedding-3-small
  • gpt-5-nano (or your preferred Responses-capable model)

Set it either as a regular environment variable:

export OPENAI_API_KEY="sk-..."

or in .env (see below).

You need an IRIS instance with:

• A namespace (in code: VECTOR)
• The ability to:
  • Create tables
  • Use VECTOR column types
  • Use VECTOR_DOT_PRODUCT and TO_VECTOR functions

The defaults in code:

• IRIS_HOST = "localhost"
• IRIS_PORT = 8881
• IRIS_NAMESPACE = "VECTOR"
• IRIS_USERNAME = "superuser"
• IRIS_PASSWORD = "sys"
• TABLE_NAME = "AIDemo.Embeddings"
• EMBEDDING_DIMENSIONS = 1536 (matches text-embedding-3-small)

Adjust these in the scripts if needed.

The scripts assume a .env file at:

• venv/dev.env

At a minimum, put your OpenAI key there:

OPENAI_API_KEY=sk-...
# Optional: override default chat model
# OPENAI_RESPONSES_MODEL=gpt-5-nano

Both embed.py and search.py load it via:

from dotenv import load_dotenv
load_dotenv(dotenv_path="venv/dev.env")

If you prefer not to use .env, just export OPENAI_API_KEY directly in your shell.

Place your Business Service documentation as .txt or .md files in:

Documentation/BusinessService/
  ├── intro.md
  ├── settings.md
  ├── examples.txt
  └── ...