🧠 Textionary

Textionary is a Streamlit-powered web application that lets you “chat” with any plain-text file. Simply upload a .txt document, ask questions in natural language, and see context-aware answers generated on the fly using Google’s Gemini LLM and LangChain.

👉 Live Demo: https://textionary.onrender.com/

---

📋 Table of Contents

1. Features
2. Tech Stack
3. Project Structure
4. Getting Started
   • Prerequisites
   • Clone & Install
   • Environment Variables
   • Run Locally
5. How It Works
6. Usage
7. Future Improvements
8. License

---

🚀 Features

• Streamlit UI: Intuitive chat interface with file uploader and message history.
• File Chunking: Splits large text into 1,000-character chunks for efficient retrieval.
• Vector Store: Persists embeddings in a Chroma DB under db_store/chroma_db_romeo_juliet/.
• Retrieval: Fetches top-3 most relevant chunks via semantic similarity.
• LLM Integration: Uses Google Gemini 2.0 (via langchain_google_genai) for both embeddings and chat generation.
• Typing Effect: Simulated “typing” animation in chat responses.
• Session State: Maintains full conversation history across interactions.
• Celebration: Streams balloons on each successful assistant response.

---

🛠️ Tech Stack

• Frontend / App Framework: Streamlit
• LLM & Embeddings: Google Generative AI (Gemini 2.0 Flash + embeddings-001)
• Orchestration: LangChain (document loaders, text splitters, prompt templates)
• Vector Database: Chroma (via langchain_chroma)
• Environment: Python 3.x, .env for secrets
• Deployment: Render (https://textionary.onrender.com/)

---

📁 Project Structure

.
├── Custom Chat.py         # Streamlit app entrypoint
├── utils.py               # Vector-store creation & retrieval functions
├── db_store/              # Persists Chroma embeddings
│   └── chroma_db_romeo_juliet/
├── requirements.txt       # Python dependencies
├── .env.example           # Example env vars (see below)
└── README.md              # This file

---

⚙️ Getting Started

Prerequisites

• Python 3.8+
• A Google Cloud project with access to the Generative AI API (Gemini)
• Render account (if you want to redeploy)

Clone & Install

git clone https://github.com/18YashDB10/Textionary.git
cd Textionary
python -m venv venv
source venv/bin/activate        # macOS/Linux
venv\Scripts\activate           # Windows
pip install -r requirements.txt

Environment Variables

Create & Copy to .env and fill in your credentials:

GOOGLE_API_KEY=your_google_api_key_here
GOOGLE_API_PROJECT=your_google_project_id

Streamlit will auto-load .env via python-dotenv.

Run Locally

streamlit run "Custom Chat.py"

Then open http://localhost:8501 in your browser.

---

🤖 How It Works

1. Upload: User uploads a .txt file via st.file_uploader.
2. Persist: The file is saved to a temporary path.
3. Embed & Store:
   • utils.create_vector_store(tmp_file_path)
   • Loads text with TextLoader
   • Splits into 1,000-character chunks (CharacterTextSplitter)
   • Embeds via GoogleGenerativeAIEmbeddings
   • Persists in a Chroma DB folder.
4. Retrieve:
   • On each query, utils.get_similar_docs(query) loads the same DB, retrieves the top 3 chunks.
5. Prompting:
   • The app wraps retrieved chunks and user question into a prompt via ChatPromptTemplate.
6. Generate:
   • Sends prompt to Gemini 2.0 Flash (ChatGoogleGenerativeAI), returns a text answer.
7. Display:
   • Shows the answer word-by-word with a typing effect, then “balloons” celebration.
   • All messages are stored in st.session_state.messages for chat history.

---

💬 Usage

1. Upload your .txt file (up to a few MB).
2. Type a question in the “What is up?” input box.
3. Read the assistant’s streamed answer.
4. Repeat as needed—your past conversation remains visible.

Example queries:

• “Summarize the main arguments.”
• “What are the key dates mentioned?”
• “Find all occurrences of the word ‘character.’”

---

🔮 Future Improvements

• 📄 Support for PDF, DOCX, and other formats
• 🔍 Adjustable number of returned chunks (k parameter)
• 🔐 User authentication and per-user DB isolation
• 📱 Responsive UI for mobile devices
• 📂 Multi‐file chat & history export

---

© 2025 Yash Deepak Bambore · GitHub Repository