LegalSphere

Empowering people to navigate the legal system with AI-powered simplicity.

---

Problem Statement

Developing an AI bot with a Voice assistant that listens to end-user questions/queries and replies back with a proper response(S2S) system. The challenge here would be that the end user can speak in any language and the bot should be able to respond in the same language. The AI bot should have a character and a back story (For example, A rude banker who hesitates to answer the query to the customer or a soft and humble actor who loves to respond to his fans) and should stick to it. The bot should answer only related to its backstory and character.

---

Proposed Solution

LegalSphere is an AI-powered, voice-interactive legal assistant that simplifies legal information using Retrieval-Augmented Generation (RAG) technology. With the Sarvam AI integration, users can engage with the platform through voice-to-voice interaction, making legal guidance more intuitive and accessible.

Key Features:

• Voice-to-Voice Interaction: Powered by Sarvam AI, users can speak their queries and receive verbal legal guidance.
• Multilingual Support: Provides legal assistance in multiple languages to serve diverse communities.
• Image Analysis: Users can upload images (e.g., legal notices or evidence) for automated analysis and legal insights.
• User-Friendly Interface: Simplifies legal complexities into easy-to-understand responses with actionable steps.
• Case Filing Assistance: Guides users through the complaint filing process, offering step-by-step legal support.

---

Technical Implementation

1. Voice Input and Output (Sarvam AI):

   • Users interact with the system using voice.
   • Sarvam AI handles speech-to-text (STT) and text-to-speech (TTS) for real-time bi-directional conversation.

2. Image Upload and Analysis:

   • Users upload images related to harassment or legal documents.
   • Gemini Pro generates a descriptive summary for analysis.

3. Semantic Understanding:

   • Text input (from voice or image) is converted into vector embeddings using BERT for deeper semantic understanding.

4. Legal Information Retrieval:

   • DeepSeek handles knowledge retrieval using RAG to query a legal database and generates context-aware responses.

5. Guidance and Support:

   • Connects users to the proper resources like NGO's and Lawyers.

---

Work Flow

---

Tech Stack

Frontend

Flutter

Built using Flutter and Dart for a cross-platform and user-friendly experience.

---

AI Models

Gemini VQA Visual Question Answering (VQA) for image interpretation

Sarvam AI Real-time voice input/output (STT & TTS)

DeepSeek Advanced deep learning techniques for knowledge retrieval and synthesis

Google Translate Google Translate for seamless translation from native languages to english

---

Architecture

1. User Input: Accepts voice, text, and images.
2. Sarvam AI: Converts speech-to-text (STT) and text-to-speech (TTS) for interaction.
3. Gemini Pro: Analyzes images for legal context.
4. BERT: Generates embeddings from text input for deep semantic understanding.
5. DeepSeek: Retrieves relevant legal information from a JSON database, Processes and refines legal responses.
6. Output: Provides legal assistance via text and voice.

---

Contributors

Sajeev Senthil

Charuvarthan

Suganth

Abiruth

Siva Prasanth Sivaraj

---

Challenges Faced

• Integrating Sarvam AI for seamless voice-based legal assistance.
• Ensuring real-time speech-to-text and text-to-speech accuracy.
• Managing secure API communications for sensitive legal queries.

---

Screen Grab

Multilingual Screen

VQA

Voice Assisstant

---

🏅 Achievement

3rd Prize Winner – GenderTech × GenAI Hackathon

Recognized for innovation in inclusive AI design and multimodal tech solutions.

• 👥 500+ teams participated across diverse domains
• 🚀 Our team made it to the Top 10 finalists
• 🥉 Secured 3rd place overall, standing out for our agentic AI approach and impact-driven execution