K2-k2/Vellum

📖 Vellum: A Soulful AI Reflection Diary

Vellum is not just a chat app; it is a digital sanctuary for your thoughts. Built with Streamlit, SQLAlchemy, and Google Gemini 2.5 Flash-Lite, Vellum acts as a poetic, empathetic companion that helps users process their emotions through structured, aesthetic journaling.

🌟 The Philosophy

Vellum is designed to feel like an old parchment diary. It uses a custom-styled interface with serif typography and soft tones to encourage deep reflection. Unlike standard assistants, Vellum is programmed to be soulful, gentle, and non-robotic.

🛠️ Key Features

• —Multimodal Journaling: Attach photos to your entries. Vellum "sees" your images and incorporates them into its poetic reflections.
• —Semantic Memory Search: Uses gemini-embedding-001 to find past moments based on meaning. Searching for "nature" will find entries about hiking even if the word "nature" wasn't used.
• —Mood Analytics: Automatically analyzes the sentiment of every entry to provide a visual donut chart of your emotional journey over time.
• —Persistent SQLite Storage: Every "whisper" is etched into a local database using SQLAlchemy, ensuring your memories survive app restarts.
• —Privacy-First Backup: Export your entire diary as a clean .txt file at any time.
• —Session Management: Built-in entry limits per session to encourage mindful journaling rather than endless scrolling.

🚀 Deployment on Hugging Face Spaces

1. 1.SDK: Select Streamlit.
2. 2.Secret Keys: Add your Gemini API key in Settings > Variables and Secrets as GOOGLE_API_KEY_N1.
3. 3.Requirements: Ensure your requirements.txt includes:

    streamlit
    pandas
    numpy
    google-generativeai
    sqlalchemy
    pillow

1. 1.Hardware: This app runs efficiently on the free CPU tier as it utilizes the lightweight Gemini 2.5 Flash-Lite model.

🏗️ Technical Architecture

• —Orchestration: Python & Streamlit
• —LLM & Vision: Google Gemini 2.5 Flash-Lite (System-prompted for a "poetic" persona)
• —Database: SQLAlchemy + SQLite (Local file-based storage)
• —Vector Search: Cosine similarity via Gemini Embeddings
• —Visualization: Vega-Lite (via Streamlit) for Mood Analytics

💻 Local Setup

# 1. Clone the repository
git clone <https://k2-k2-vellum.hf.space>

# 2. Install dependencies
pip install -r requirements.txt

# 3. Set your API Key
# Windows: set GOOGLE_API_KEY_N1=your_key
# Mac/Linux: export GOOGLE_API_KEY_N1="your_key"

# 4. Run Vellum
streamlit run app.py

Developed with ❤️ by Kesar Agarwal

Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference