🏅 ParaBot: AI-Driven Accessible Voice Assistant

ParaBot is an advanced AI assistant designed to eliminate bureaucratic and digital barriers for Paralympic athletes. It transforms complex regulations (PDFs) into a fluid, 100% hands-free voice conversation.

Developed as the Capstone Project for the Samsung Innovation Campus, it received a 9.9/10 grade for its technical excellence and impact on digital inclusion.

---

🏗️ Technical Architecture

ParaBot is built on a robust RAG (Retrieval-Augmented Generation) architecture, ensuring that answers are grounded in official documentation and reducing model hallucinations.

• IA Engine: Orchestrated with Llama 3 (via Groq for low-latency inference).
• Embeddings: paraphrase-multilingual-MiniLM-L12-v2 from Sentence-Transformers.
• Vector Database: FAISS for efficient semantic retrieval.
• Multimodal Pipeline:
  • Speech-to-Text (STT): OpenAI Whisper via Groq.
  • Text-to-Speech (TTS): gTTS for natural response narration.
• Frontend: Streamlit with custom CSS for high-contrast and keyboard navigation.

---

♿ Accessibility First (WCAG 2.1 AA+)

True engineering is inclusive. ParaBot was designed from the ground up to meet the WCAG 2.1 Level AA+ standards:

• Visible High-Contrast Focus: All interactive elements are keyboard-navigable with clear focus indicators.
• Dark Mode Optimization: Intelligent adaptation to user preferences for better readability.
• Easy Reading Mode: A specialized feature that uses Prompt Engineering to simplify complex legal/technical language in real-time.
• Hands-Free Interaction: Full integration of voice recording and auto-reading of responses.

---

🚀 Getting Started

Prerequisites

• Python 3.12+
• A Groq API Key

Installation

1. Clone the repository:

   git clone https://github.com/your-username/ParaBot.git
   cd ParaBot

2. Install dependencies:

   pip install -r requirements.txt

3. Set up your environment:

   • Copy .env.example to .env.
   • Add your GROQ_API_KEY and a password for the admin panel.

4. Run the application:

   streamlit run app.py

---

🛠️ Admin Panel

Authorized users can upload new PDFs to the ./pdfs folder through the secure admin panel. The system automatically handles chunking, embedding generation, and FAISS index updates without manual intervention.

---

📜 Project Context

This project was part of the Samsung Innovation Campus program at Universitat Politècnica de València (UPV).

• Grade: 9.9 / 10
• Focus: AI Ethics, Digital Inclusion, and Software Engineering.

---

⚖️ License

Distributed under the MIT License. See LICENSE for more information.

---

Created with 🏅 and ☕ by Sergi Torres Albert