SAFARSETU: An AI-Powered Multilingual Tourist Guide Chatbot for Cultural Heritage Exploration

Abstract

India’s architectural heritage, particularly the vernacular Havelis (traditional mansions), faces a critical dual challenge: physical deterioration and a fading public narrative. While the \"Digital India\" initiative has accelerated technology adoption, existing heritage tourism solutions often rely on high-bandwidth mobile applications or English-centric interfaces, creating a \"digital divide\" that excludes a vast demographic of domestic rural tourists. This paper proposes SAFARSETU, a serverless, zero-install, multilingual chatbot built on the Telegram platform designed to democratize access to heritage information. Unlike traditional synchronous bots, our architecture employs a novel Asynchronous NLP Pipeline utilizing Python’s asyncio library to parallelize Neural Machine Translation (NMT) via Google Translator and Neural Text-to-Speech (TTS) via Edge-TTS. We introduce a custom \"Sentence-Aware Chunking Algorithm\" to overcome Telegram’s 4096-character message limit without disrupting semantic continuity or audio prosody. Experimental validation across 500+ interaction cycles demonstrates that the system maintains a Total Response Time (TRT) of <1.8 seconds for complex multilingual queries, significantly outperforming synchronous architectures. SAFARSETU offers a scalable, inclusive reference architecture for digital heritage tourism in resource-constrained environments.

Introduction

The text discusses SAFARSETU (Journey Bridge), an AI-powered, multilingual chatbot designed to enhance heritage tourism in India, focusing on accessibility, vernacular support, and low-latency delivery for users with limited connectivity. Traditional heritage preservation efforts are limited to physical restoration or high-end digital solutions that often exclude rural tourists due to device, bandwidth, or language constraints. SAFARSETU addresses these gaps by leveraging the Telegram platform, asynchronous processing, and Neural Machine Translation (NMT) with Neural Text-to-Speech (TTS) to provide an inclusive, immersive experience.

Key Points:

1. Problem Context:

   • Indian Havelis face physical decay and declining public engagement due to lack of accessible digital narratives.

   • Existing digital solutions are often high-bandwidth, English-centric, and visually focused, creating a “digital divide” for rural and low-literacy tourists.

   • Current heritage chatbots are static, synchronous, and lack vernacular support or audio storytelling.

2. Technological Innovation (SAFARSETU):

   • Platform Choice: Telegram chosen for low-latency performance, data compression, and accessibility in 2G/EDGE networks.

   • Asynchronous NLP Pipeline: Uses Python’s asyncio to parallelize NMT and TTS, reducing perceived latency by over 70% compared to synchronous systems.

   • Semantic Chunking Algorithm: Segments long historical texts for Telegram’s 4096-character limit without disrupting sentence flow, improving TTS prosody.

   • Neural Audio & Vernacular Support: High-quality voice synthesis in English, Hindi, and Urdu, accommodating visually impaired and low-literacy users.

   • Serverless, Event-Driven Architecture: Microservices-based system offloads computation to backend, ensuring high concurrency and low operational costs.

3. System Design & Implementation:

   • Data Layer: Read-optimized JSON repository loaded in-memory for fast retrieval.

   • Controller & Logic Layers: Webhook-based Telegram interface with asynchronous task orchestration for concurrent translation, audio generation, and visual content delivery.

   • Sanitization & Audio Mapping: MarkdownV2 sanitization and phonetic dictionaries enhance TTS reliability and pronunciation of heritage-specific terms.

4. Evaluation & Results:

   • Laboratory Benchmarking: Asynchronous pipeline reduced total response time by 22.5% and perceived latency (Time to First Byte) by 74.3% compared to synchronous baselines.

   • Stress Testing: Maintained 99.2% success under 100 concurrent users, whereas synchronous bots failed 18% of the time.

   • Field Pilot Study: Conducted at the Residency Complex, Lucknow with 120 participants; users preferred audio-first interaction (78%), switched to local languages quickly (62%), and rated usability, usefulness, and reliability highly (scores 4.55–4.88/5).

5. Impact & Contributions:

   • SAFARSETU bridges the digital divide by providing low-bandwidth, vernacular, and multimodal access to heritage narratives.

   • Enhances user engagement through fast, asynchronous delivery and immersive audio storytelling.

   • Demonstrates scalability and inclusivity in real-world heritage tourism contexts, outperforming prior chatbots like "TN Forts Buddy" in latency, concurrency, and user satisfaction.

Conclusion

This research presented SAFARSETU, an AI-powered multilingual conversational agent designed to democratize access to India’s vernacular architectural heritage. By addressing the critical limitations of existing digital solutions—specifically their reliance on high-bandwidth applications, English-centric interfaces, and synchronous processing models—SAFARSETU establishes a new benchmark for accessible digital tourism. The study yields three significant conclusions. First, the successful deployment of the Asynchronous NLP Pipeline demonstrates that serverless, event-driven architectures can effectively handle the \"bursty\" traffic patterns of tourism without the latency bottlenecks observed in traditional synchronous bots [9]. By decoupling content retrieval from audio synthesis, our system achieved a 22.5% reduction in Total Response Time, ensuring a seamless user experience even on 2G networks. Second, the integration of Neural Text-to-Speech (TTS) with vernacular support (Hindi/Urdu) directly bridges the \"Digital Divide.\" Our field data, which showed a 78% user preference for audio narratives, validates the argument by Benaddi et al. that voice-driven interfaces are essential for inclusive heritage exploration [6]. This \"Audio-First\" approach transforms heritage sites from visual spectacles into immersive storytelling environments, making history accessible to the visually impaired and linguistically diverse populations [14]. Finally, the adoption of the Telegram platform as a delivery mechanism proves that sophisticated cultural preservation does not require expensive, heavy mobile applications. As noted by Nafis et al., leveraging ubiquitous messaging platforms provides \"real added value\" by lowering the barrier to entry for domestic tourists [4]. SAFARSETU effectively combines the depth of historical knowledge found in archival systems with the accessibility of a chat interface. While SAFARSETU offers a robust foundation, we identify several avenues for future research and enhancement: Expansion to Dialects: Currently, the system supports standard Hindi and Urdu. Future iterations will integrate models for regional dialects such as Bhojpuri, Marwari, and Awadhi. Resemble AI (2024) highlights that hyper-local accents are crucial for establishing deeper cultural trust and authenticity [8]. Multimodal \"Snap-to-Ask\" Functionality: We plan to integrate Computer Vision capabilities (e.g., OpenAI CLIP or Google Lens APIs) to allow users to upload a photograph of a specific architectural feature (e.g., a Jharokha) and receive an instant historical explanation. This aligns with the multimodal trends identified by Deng in the evolution of experience design [5]. Community-Driven Content Curation: To scale the .json repository beyond Lucknow, we propose a \"Wiki-Heritage\" web portal where local historians and guides can submit and verify new monuments. This crowdsourced approach, supported by verification algorithms, will ensure the long-term sustainability and accuracy of the knowledge base [11]. Offline Caching: Recognizing the unstable connectivity in remote heritage belts, we intend to implement a \"Lite Mode\" that caches textual data and lightweight audio maps onto the user\'s device for offline access [12]. In conclusion, SAFARSETU serves as a scalable, technically rigorous, and socially inclusive reference architecture for the digital preservation of tangible heritage in the Global South.

References

[1] K. Sathiyabamavathy and K. P. Anju, \"Role of Chatbots in Cultural Heritage Tourism: An Empirical Study on Ancient Forts and Palaces,\" Journal of Heritage Management, vol. 9, no. 1, pp. 9–28, 2024. [2] D. Deepa, A. K. Archana, and K. Karthik, \"Heritage Information Chatbot,\" International Journal of Emerging Technologies and Innovative Research (JETIR), vol. 12, no. 7, pp. 290–293, July 2025. [3] P. Reddy and A. Kumar, \"Enhancing Visitor Experience Through a Chatbot for Historical Places in India Using Google Dialogflow,\" Journal of Engineering Sciences, vol. 15, no. 4, pp. 222–230, 2024. [4] F. Nafis, A. Yahyaouy, and B. Aghoutane, \"Chatbots for Cultural Heritage: A Real Added Value,\" in Proceedings of the 2nd International Conference on Big Data, Modelling and Machine Learning (BML), 2021, pp. 502–506. [5] M. Deng, \"Machine Learning Advances in Technology Applications: Cultural Heritage Tourism Trends in Experience Design,\" International Journal of Advanced Computer Science and Applications (IJACSA), vol. 16, no. 4, pp. 186–196, 2025. [6] Benaddi et al., \"Voice-Driven Heritage Tour Assistance System using NLP and Geolocation,\" International Journal of Innovative Research in Science, Engineering and Technology (IJIRSET), vol. 14, no. 3, pp. 415–422, Mar. 2025. [7] R. Boboc, E. B?utu, and F. Gîrbacia, \"Augmented Reality and AI in Cultural Heritage: An Overview of the Last Decade,\" Applied Sciences, vol. 12, no. 19, p. 9859, 2022. [8] Resemble AI, \"The Importance of Regional Accents in AI Voice Generators for Cultural Preservation,\" AI Ethics Report, 2024. [Online]. Available: https://www.resemble.ai/indian-accent-text-to-speech. [9] J. Hu et al., \"DeepServe: Serverless Large Language Model Serving at Scale,\" in Proceedings of the USENIX Annual Technical Conference (ATC), Boston, MA, USA, July 2025. [10] S. Alekseev et al., \"Telegram Bot Development Using Python: An Educational Architecture,\" International Journal of Emerging Technologies, vol. 11, no. 4, pp. 30–35, 2024. [11] T. Le and C. Arcadia, \"Proposing a Systematic Approach for Integrating Traditional Research Methods into Machine Learning in Hospitality,\" Current Issues in Tourism, vol. 24, no. 12, pp. 1640–1655, 2021. [12] T. K. Gireesh Kumar, \"A Study on Digital Preservation Methods for Cultural Heritage Sites in India,\" Asian Journal of Information Science and Technology, vol. 14, no. 2, pp. 45–52, 2024. [13] Ministry of Tourism, Government of India, \"India Tourism Statistics 2024,\" Market Research Division, New Delhi, 2024. [14] D. Harisanty et al., \"Cultural Heritage Preservation in the Digital Age: Harnessing Artificial Intelligence,\" Digital Library Perspectives, vol. 40, no. 4, pp. 609–625, 2024. [15] [M. Casillo, F. Clarizia, and D. Santaniello, \"CHAT-Bot: A Cultural Heritage Aware Teller-Bot for Supporting Touristic Experiences,\" Pattern Recognition Letters, vol. 131, pp. 234–243, 2020.

Copyright

Copyright © 2025 Anurag Shrivastava , Shivang Agrawal, Sanjana Keshari, Krishna Vishwakarma, Mohd. Taukeer. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.