Store Intelligence Assistant

An interactive Streamlit app for exploring store performance, inventory, and business policies, powered by Databricks and OpenAI agents SDK.

What is this app?

This app is a conversational assistant that helps users query and analyze:

• 📊 Store performance metrics
• 🔄 Return and business conduct policies
• 🛍️ BOPIS (Buy Online, Pick Up In Store) information
• 📦 Product inventory across stores
• 📈 Monthly Sales Forecasts for stores

It leverages Databricks-hosted resources, Databricks Genie rooms, and a large language model (LLM) to provide on-demand, context-aware answers. The repo also shows how to deploy this app to Databricks Apps.

---

Quick Demo

The following is a quick demo of the working application:

Key Databricks Resources

1. Databricks Genie Rooms

The app uses Genie rooms to power its data tools. These Genie rooms are used by the app's tools to fetch live data and answer user questions.

• Store Performance Genie Room: For revenue, BOPIS, forecasts, store performance metrics

• Product Inventory Genie Room: For inventory snapshots, product details etc.

2. Databricks Vector Search

• Business Conduct Policy Table: For querying and finding out specific information about vendor conduct policy using a vector index

3. LLM (Large Language Model)

• The app uses a Databricks-hosted LLM, specified by the DATABRICKS_MODEL environment variable.
• The LLM is accessed via the OpenAI-compatible API endpoint provided by Databricks Mosaic AI Gateway.
• All chat and tool responses are generated or orchestrated by this model.

---

How does it work?

• The app is built with Streamlit for a modern, interactive UI.
• User questions are routed to an agent, which can:
  • Answer directly using the LLM
  • Call specialized tools that query Genie rooms or Databricks tables for live data
• All Databricks credentials and resource IDs are managed via environment variables (see .env setup).

---

Tools & Data Flows

• get_store_performance_info:
  • Queries the Store Performance Genie room for sales, location, and forecast data.
• get_product_inventory_info:
  • Queries the Product Inventory Genie room for inventory snapshots and product data.
• get_business_conduct_policy_info:
  • Queries a Databricks Vector Search Index for business conduct and return policies.

---

Vector Search Index

The vector search index was hydrated using the process highlighted in the vector_search_setup.png file in this repository.

---

Quickstart

1. Clone this repo and install dependencies:

   pip install -r requirements.txt

2. Set up your .env file with Databricks credentials and model info.
3. Run the app:

   streamlit run streamlit_multi_genie_tools.py

---

Customization

• Update the Genie room IDs or Databricks model in your .env file to point to your own resources.
• The UI and agent instructions can be customized in streamlit_multi_genie_tools.py.

---

Requirements

• Python 3.12
• Access to a Databricks workspace with Genie enabled
• A Databricks-hosted LLM (or compatible OpenAI endpoint)

To Run the App Locally

1. Set up a .env file with the following details note: you only need the MLFLOW_EXPERIMENT_ID if you want to enable montioring or write to an existing external monitor

DATABRICKS_TOKEN=
DATABRICKS_HOST=
DATABRICKS_BASE_URL=
DATABRICKS_MODEL=
GENIE_SPACE_ID=
GENIE_SPACE_PRODUCT_INV_ID='
MLFLOW_EXPERIMENT_ID=

2. Create a virtual env locally using uv uv venv --python 3.12
3. Then, install the requirements using uv pip install -r requirements.txt
4. Run the app streamlit run streamlit_multi_genie_tools.py
5. You might have to create a .streamlit folder on the root of the project to control the theme as you might see fit using a config.toml file

Example config.toml file

[theme]
base="light"
primaryColor="#FF4B4B"
backgroundColor="#FFFFFF"
secondaryBackgroundColor="#F0F2F6"
textColor="#31333F"
font="sans serif"

---

Deploying to Databricks Apps

You can deploy this Streamlit app to your Databricks workspace as a Databricks App for secure, scalable access. Here's how:

1. Prepare app.yaml and Secrets

• The app uses an app.yaml file to define how it runs and which environment variables it needs.
• Best practice: Use Databricks secrets to store sensitive values (tokens, workspace URLs, model IDs, etc).
• In your Databricks workspace, add secrets for each required variable (see below).

Example app.yaml:

command: ['streamlit', 'run', 'streamlit_multi_genie_tools.py']
env:
  - name: 'DATABRICKS_TOKEN'
    valueFrom: 'token'  # Secret name in Databricks
  - name: 'DATABRICKS_HOST'
    valueFrom: 'DATABRICKS_HOST'
  - name: 'DATABRICKS_BASE_URL'
    valueFrom: 'DATABRICKS_BASE_URL'
  - name: 'DATABRICKS_MODEL'
    valueFrom: 'DATABRICKS_MODEL'
  - name: 'GENIE_SPACE_ID'
    valueFrom: 'GENIE_SPACE_ID'
  - name: 'GENIE_SPACE_PRODUCT_INV_ID'
    valueFrom: 'GENIE_SPACE_PRODUCT_INV_ID'
  - name: 'MLFLOW_EXPERIMENT_ID'
    valueFrom: 'MLFLOW_EXPERIMENT_ID'

• Each valueFrom should match the name of a secret you've configured in your Databricks workspace (see Databricks docs).

2. Create the App in Your Workspace

In your terminal, run:

databricks apps create <app-name>

Replace <app-name> with your desired app name (e.g., store-intel-assistant).

3. Sync Your Local App to Databricks Workspace

From your project directory, run:

databricks sync --watch . /Workspace/Users/<your-user-or-group>/<app-name>

• This will upload your code and watch for changes.
• Exclude files/folders using .gitignore if needed.

4. Deploy the App

After syncing, deploy the app with:

databricks apps deploy <app-name> --source-code-path /Workspace/Users/<your-user-or-group>/<app-name>

• Replace <app-name> and the workspace path as appropriate.

5. View and Use the App

• Go to your Databricks workspace, click Compute > Apps tab, and find your app.
• Click the app name to view deployment status and launch the Streamlit UI.

---

References:

• Databricks Apps: Get Started
• Databricks Apps: App Development & Secrets
• Databricks Apps: Configuration