Smart Search with GenAI

An intelligent, LLM-powered semantic search tool for querying structured data using natural language.

---

🔍 Overview

Smart Search with GenAI is a semantic search application that combines vector similarity search (FAISS) with Azure OpenAI's powerful LLMs to answer user queries over structured datasets. Users can input natural language questions and receive context-aware, summarized responses based on pre-indexed data.

Built with Streamlit, the UI offers an interactive, user-friendly experience with prompt suggestions, onboarding guidance, and instant query answers.

---

✨ Features

• 🔎 Natural language search over structured data (Excel)
• 🧠 Azure OpenAI embedding + summarization (text-embedding-ada-002 + gpt-4o-mini)
• ⚡ Fast retrieval using FAISS vector index
• 🔁 Chunk reranking using cosine similarity for better accuracy
• 💡 Suggested prompt buttons to guide user input
• 👋 Onboarding walkthrough for new users
• 📊 Built-in Streamlit interface (no frontend coding needed)

---

Solution Architecture:

---

🧱 Tech Stack

Category	Tools Used
UI	Streamlit + Custom CSS
Embedding Model	Azure OpenAI text-embedding-ada-002
Completion Model	Azure OpenAI gpt-4o-mini (configurable)
Vector DB	FAISS
Reranking	Cosine Similarity (Scikit-learn)
Data Format	Excel (.xlsx) + Pandas
Configuration	Python dotenv

---

📦 Project Structure

smart_search/
├── README.md
├── app.py                         # Streamlit UI
├── pyproject.toml                 # Project metadata & dependencies
├── .python-version                # Python version spec
├── uv.lock                        # Dependency lock file (if committed)
├── .gitignore
│
├── .env                          # [Not committed] Environment variables
│
├── data/                         # Input data
│   ├── raw/
│   │   ├── people_data_100.xlsx
│   │   ├── people_data_500.xlsx
│   │   └── people_data_1000.xlsx
│   ├── logo/
│   │   ├── smart_search_logo.png
│   │   └── favicon.ico
│   └── processed/
│       └── people_data_1000_with_textchunk.xlsx
│
├── embeddings/                   # Vector storage & metadata
│   ├── chunk_embeddings.npy
│   ├── faiss_index_people_data.index
│   └── metadata.csv
│
├── prompts/
│   └── prompt_v1.txt             # LLM prompt template
│
├── scripts/                      # One-time/utility scripts
│   ├── __init__.py
│   ├── embed_and_index.py        # Embedding + indexing pipeline
│   ├── preprocess.py             # Excel data transformation
│   ├── search.py                 # Basic FAISS query CLI
│   └── search_with_llm.py        # RAG CLI interface
│
├── styles/
│   ├── base.css              # Global variables, resets
│   ├── theme.css             # Light/Dark theme styling
│   ├── buttons.css           # Button design + animation
│   ├── loader.css            # Custom loader animation
│   ├── animations.css        # Additional animations (e.g., onboarding)
│   └── style.css             # (Optional) Imports all above files if needed
│
│
└── utils/                        # Reusable backend logic
    ├── __init__.py
    ├── answer_generator.py       # Prompt & LLM answer generator
    ├── azure_openai_client.py    # Auth wrapper for Azure OpenAI
    ├── config.py                 # Env & path configs
    ├── embedder.py               # Query embedder
    ├── examples.py               # Suggested prompt examples
    ├── onboarding.py             # First-time user walkthrough
    ├── prompt_loader.py          # Load prompt from file
    ├── theme.py                  # # Theme CSS injection
    ├── reranker.py               # Cosine-based reranker
    └── search_engine.py          # Semantic + reranked search logic

---

Data Flow Diagram:

               ┌─────────────────────────────┐
               │      🧾 Excel Dataset       │
               └──────────────┬──────────────┘
                              │
                              ▼
               ┌─────────────────────────────┐
               │ Data Preprocessing (Python) │
               └──────────────┬──────────────┘
                              ▼
     ┌─────────────────────────────────────────────────────┐
     │   TextChunks created + saved to processed .xlsx     │
     └────────────────────────┬────────────────────────────┘
                              ▼
     ┌─────────────────────────────────────────────────────┐
     │       Azure OpenAI (text-embedding-ada-002)         │
     │         Embed TextChunks → 1536-dim vector          │
     └────────────────────────┬────────────────────────────┘
                              ▼
     ┌─────────────────────────────────────────────────────┐
     │ FAISS Index (in-memory vector DB)                   │
     │ Stores & retrieves similar chunks                   │
     └────────────────────────┬────────────────────────────┘
                              ▼
     ┌─────────────────────────────────────────────────────┐
     │ Streamlit UI                                        │
     │  - Input query → Embed                              │
     │  - Top-K retrieval + reranking                      │
     │  - LLM prompt creation                              │
     │  - Azure OpenAI chat completion (gpt-4o-mini)       │
     │  - Final answer rendered                            │
     └─────────────────────────────────────────────────────┘

---

⚙️ Setup Instructions

1. Clone the repository

   git clone https://github.com/ajeet214/smart_search.git
   cd smart_search

2. Install dependencies and setup environment using uv

   • Initialize environment (if not already done):

     uv init

   • Install dependencies (defined in pyproject.toml):

     uv sync

3. Activate the virtual environment

   • Using uv:

     uv shell

   • Or activate .venv manually::

     source .venv/bin/activate      # Linux/macOS
     .venv\Scripts\activate         # Windows

4. Configure environment

   Create a .env file at the root and fill the

   • AZURE_OPENAI_API_KEY,
   • AZURE_OPENAI_ENDPOINT , leave all other as it is.
   • You may change the SHOW_ONBOARDING to false if you like to disable the onboarding screen.

   # Azure OpenAI
   AZURE_OPENAI_API_KEY=your_key_here
   AZURE_OPENAI_ENDPOINT=https://your-endpoint.openai.azure.com
   AZURE_OPENAI_API_VERSION=2023-05-15
   AZURE_OPENAI_DEPLOYMENT=text-embedding-ada-002
   AZURE_OPENAI_COMPLETION_DEPLOYMENT=gpt-4o-mini
   
   # File Paths
   INPUT_FILE=data/processed/people_data_1000_with_textchunk.xlsx
   OUTPUT_INDEX=embeddings/faiss_index_people_data.index
   OUTPUT_METADATA=embeddings/metadata.csv
   CHUNK_EMBEDDINGS_PATH=embeddings/chunk_embeddings.npy
   
   # UI Behavior
   SHOW_ONBOARDING=true

---

4. Prepare Data

python scripts/preprocess.py

---

5. Generate Embeddings and Build Index

python scripts/embed_and_index.py

---

Launch the App

To start the Streamlit web app, run:

uv run streamlit run app.py

or if your virtual environment is active:

streamlit run app.py

Then open your browser and go to:

http://localhost:8501

---

💡 Example Prompts

• "Who attended AI-related events in Singapore?"
• "List members of the Data Science team based in Tokyo."
• "What events did Alice Ly take part in?"
• "Show the Marketing team members from Vietnam."

---

🔒 Security Note

This project uses Azure OpenAI via environment variables. Never commit your .env file. Always include .env in .gitignore.

---

---

🚀 Roadmap / Future Enhancements

• Support multi-file or live data ingestion
• Add citation reference to answers
• Export responses (PDF, Markdown)
• Chat history tracking
• Authentication layer for private access

---

📄 License

MIT License. Feel free to fork and extend.

---

🙌 Acknowledgements

• Streamlit
• FAISS
• Azure OpenAI

---

Let me know if you'd like me to help you add badges, deployment instructions (e.g., for Azure App Service or HuggingFace Spaces), or GitHub Actions CI/CD!

Built with ❤️ by Ajeet using Azure OpenAI, Streamlit, and FAISS.