DataToad – Conversational Business Intelligence for Sales Patterns

DataToad is a full-stack Business Intelligence application that lets business users explore sales patterns and forecasts through a conversational interface. It combines Databricks, advanced time-series signal processing (STFT, wavelets, Hilbert–Huang Transform), and an LLM hosted via NVIDIA NIM to surface actionable insights.

---

What It Is

• Conversational BI assistant — Ask natural language questions about sales patterns, trends, spikes, dips, volatility, and seasonality
• Pattern mining on Databricks — Reads precomputed pattern metadata and raw sales data from your Databricks Lakehouse
• Pattern-aware forecasting — Generates future sales projections and translates them into business language via the LLM
• Transform pipeline — Triggers a Databricks job to (re)compute pattern metadata or run a local transform pipeline for testing
• Caching — Redis-backed cache for faster repeated queries and forecasts

---

Architecture

┌─────────────────┐
│  Chat Interface  │
│    (Next.js)     │
└────────┬────────┘
         │
         ▼
┌─────────────────────────────────┐
│        FastAPI Backend           │
│  • Query / chat / patterns       │
│  • Forecast generation           │
│  • Transform pipeline            │
└──┬──────────┬──────────┬────────┘
   │          │          │
   ▼          ▼          ▼
┌────────┐ ┌───────┐ ┌──────────┐
│Databricks│ │ Redis │ │NVIDIA NIM│
│Lakehouse │ │ Cache │ │   LLM    │
└──────────┘ └───────┘ └──────────┘

• Frontend: Next.js 14 app with a single-page conversational UI
• Backend: FastAPI handles queries, forecasting, and pipeline runs
• Data layer: Databricks stores pattern metadata, sales data, and forecasts
• LLM: NVIDIA NIM for natural language summaries and explanations
• Cache: Redis for repeated queries

For details, see docs/ARCHITECTURE.md.

---

How to Use

Local setup

Prerequisites: Python 3.10+, Node.js 18+, Redis, Databricks workspace, NVIDIA NIM API key

Backend

git clone https://github.com/Pshyam17/DataToad.git
cd DataToad

python -m venv venv
source venv/bin/activate  # Windows: venv\Scripts\activate
pip install -e .

Create a .env file with Databricks, NVIDIA NIM, and Redis credentials. Run:

uvicorn src.api.main:app --reload --host 0.0.0.0 --port 8000

Frontend

cd frontend
npm install

Create frontend/.env.local with NEXT_PUBLIC_API_URL=http://localhost:8000. Run:

npm run dev

Open http://localhost:3000 in your browser.

Using the interface

• Pattern questions — e.g. Show me products trending up, Find products with spikes in the last quarter
• Forecasts — e.g. Forecast sales for Product_123 for the next 6 months
• Pipeline runs — e.g. Run the pattern analysis pipeline

Use the quick filters (Trending up, Spikes, Volatile) above the input, or type questions directly.

---

Deploy to Vercel

The frontend (Next.js) can be deployed to Vercel. The backend must be hosted separately (e.g. Railway, Render, or a container service) and its URL configured in the frontend.

1. Connect the repo

1. Go to vercel.com and sign in
2. Click Add New → Project
3. Import the Pshyam17/DataToad GitHub repository
4. Vercel will detect Next.js (the project has vercel.json with rootDirectory: 'frontend')

2. Configure build settings

• Framework preset: Next.js
• Root directory: frontend
• Build command: npm run build (default)
• Output directory: .next (default)

3. Set environment variables

In Settings → Environment Variables, add:

Variable	Value	Notes
NEXT_PUBLIC_API_URL	https://your-backend-url.com	Your deployed FastAPI backend URL

4. Deploy

• Click Deploy
• Vercel builds and deploys the frontend
• Use the provided Vercel URL (e.g. https://datatoad.vercel.app)

5. Backend requirement

The backend (FastAPI) must be deployed and reachable at the URL set in NEXT_PUBLIC_API_URL. It needs access to Databricks, Redis, and NVIDIA NIM. Deploy it to Railway, Render, or a similar service, then update the environment variable and redeploy the Vercel project if needed.