The problem Raasta - Bridging the gap of languages and travel solves

In bustling cities, such as those in India, transportation apps encounter challenges in providing quick and effective solutions due to diverse languages spoken by the users. Moreover, there is a significant issue related to transparent cost estimates for individuals with budget constraints.
Our Solution 📱:
Raasta employs a diverse set of AI tools to effectively eliminate linguistic and economic barriers in the realm of urban mobility. With this innovative solution, users can effortlessly access real-time traffic advisories in their preferred language. Furthermore, Raasta goes a step further by providing personalized route suggestions based on the user's budget, ensuring a seamless and cost-effective travel experience.
Key Features 🤖:

1. Enhancing Urban Mobility Accessibility:
   Deployment of Indic Translation Tools to bridge language gaps.
   Introduction of Voice-Based Interfaces tailored for individuals with disabilities and the elderly, simplifying interactions with the transportation system.
2. Promoting Cost-Effective Urban Mobility: LAX.ai
   Implementation of Lax.ai Solutions, offering budget-conscious route planning.
   Intelligent route recommendations based on various factors, including weather conditions, to optimize costs for users.
3. Simplifying Urban Mobility Procedures:
   Introduction of a Challan Summarizer and Extractor to simplify traffic violation procedures.(OCR AND PDF MINER)
   Integration of the RAG model for rule checking, ensuring adherence to traffic regulations from the knowledge base (used MISTRAL 7B with embedchain)
4. Localization of procedures in Indic languages:
   Providing users, the option to navigate transportation processes in their preferred languages for enhanced simplicity and understanding. (USING INDICTRANS2 OF AIFORBHARAT).

Challenges we ran into

1. Integration Complexities:
   Incorporating diverse AI tools to the React Frontend posed integration complexities.
   Solution: Iterative testing and collaboration resolved integration issues.
2. Language Model Fine-tuning:
   Fine-tuning RAG models like Llama2 for dataset efficiency.
   Solution: Implemented PEFT methods like Qlora for optimal VRAM efficiency.
3. Challan Extraction Accuracy:
   Achieving high accuracy in OCR-based challan extraction.
   Solution: Experimented with pytesseract and easyocr parameters and fine-tuned OCR models for improved accuracy.
   Multilingual Speech-to-Speech:
4. Implementing Speech-to-Speech for 22 Indic languages.
   Solution: Experimented with Distilled Whisper LLM, gTTS, and PLAYHT, optimizing for diverse linguistic nuances.
   Dynamic Language Support:
5. Expanding language support for the chatbot.
   Solution: Integrated AI4Bharat's INDICTRANS2 LLM and continuously refined language models for broader coverage.
   Overcoming these challenges involved a combination of experimentation, collaboration, and iterative development, leading to a robust and functional urban mobility solution. 🛠️💡