Vanna with NVIDIA AI Endpoints (#290)

* Vanna with NV

Signed-off-by: Divyansh Jain <[email protected]>

* Moving vanna to community folder

* Remove Vanna files from langchain folder

* Add README and project assets

---------

Signed-off-by: Divyansh Jain <[email protected]>

Author: divyanshjain

community/README.md

@@ -23,6 +23,10 @@ Community examples are sample code and deployments for RAG pipelines that are no
 
 ## Inventory
 
+* [Vanna with NVIDIA AI Endpoints](./Vanna_with_NVIDIA_AI_Endpoints/)
+
+  This example demonstrates how to optimize Vanna's open-source text-to-SQL pipeline using NVIDIA NIM and NeMo Retriever for faster and more efficient analytics. The notebook showcases integration with NVIDIA's AI endpoints to accelerate natural language to SQL queries on a Steam games dataset, enabling data exploration through simple English questions that are automatically converted to optimized SQL queries.
+
 * [NVIDIA Data Analysis Agent](./data-analysis-agent/)
 
   This example demonstrates an interactive, agentic data analysis application that leverages NVIDIA Llama-3.1-Nemotron-Ultra-253B-v1 for advanced reasoning and data exploration. Users can upload CSV files, ask questions in natural language, and receive automated visualizations with clear, step-by-step reasoning. The implementation features a modular agent architecture for data insight, code generation, execution, and transparent reasoning.

community/Vanna_with_NVIDIA_AI_Endpoints/README.md

@@ -0,0 +1,68 @@
+# Accelerating Text-to-SQL Inference with Vanna and NVIDIA AI Endpoints
+
+This repository demonstrates how to optimize Vanna's open-source text-to-SQL pipeline using NVIDIA NIM and NeMo Retriever for faster and more efficient analytics.
+
+![NVIDIA Accelerated Vanna Pipeline](assets/nvidia_vanna.png)
+
+## Overview
+
+This project showcases how to leverage NVIDIA's AI endpoints to significantly accelerate text-to-SQL generation with Vanna, enabling faster natural language queries on databases.
+
+## Features
+
+- Integration with NVIDIA NIM for LLM-based SQL generation
+- NeMo Retriever for efficient semantic search
+- Training Vanna on domain specific data
+
+## Requirements
+
+- Python 3.10+
+- NVIDIA API key (get yours at [https://build.nvidia.com/](https://build.nvidia.com/))
+- Required Python packages:
+  - vanna
+  - langchain
+  - langchain-nvidia-ai-endpoints
+  - pymilvus[model]
+  - pandas
+  - numpy
+  - sqlite3
+  - openai
+  - kagglehub
+
+## Getting Started
+
+1. Clone this repository
+2. Install the required packages
+3. Set your NVIDIA API key
+4. Run the Jupyter notebook `vanna_with_NVIDIA.ipynb`
+
+## Dataset
+
+The notebook uses Steam game datasets from Kaggle:
+- [Steam Games Dataset](https://www.kaggle.com/datasets/fronkongames/steam-games-dataset/data)
+- [Games on Steam](https://www.kaggle.com/datasets/sujaykapadnis/games-on-steam/)
+
+After preprocessing, the data is organized into three tables:
+- `games`: Game information including name, price, reviews, etc.
+- `categories`: Game categories
+- `tags`: User-defined tags and their frequencies
+
+## Pipeline
+
+The NVIDIA-accelerated Vanna pipeline consists of:
+1. Data preparation and ingestion
+2. Vector DB setup with Milvus
+3. Integration with NVIDIA LLM (Llama 3.1) and embedding models
+4. SQL database setup and training
+5. Natural language query to SQL conversion
+
+## Example Queries
+
+The notebook demonstrates several example queries:
+- "Which 5 games have the most positive reviews and how many?"
+- "Which indie game has the biggest player base?"
+- "Which category has the maximum number of gamers and how many?"
+
+## License
+
+This project is licensed under the MIT License - see the LICENSE file for details.