SafarSaathi: A Real-Time Multilingual Public Transport Companion Delivered over WhatsApp

Author

Sangam Mishra, Avadhut Kaskar, Shubham Partole, Prof. Satish Kuchiwale


Abstract

 Commuters in Indian cities make millions of bus, metro and suburban rail trips every day, yet live transit information mostly sits inside map apps that expect a smartphone, mobile data, English literacy and some comfort with maps. This paper presents SafarSaathi, a multilingual public-transport companion that runs entirely inside WhatsApp. A commuter sends a text message, a voice note or a location pin in Hindi, Marathi or English. A large language model (LLM) layer identifies the language, the intent, and the origin and destination. OpenTripPlanner then plans a multimodal route over open GTFS schedules merged with GTFS-Realtime updates, and the answer (the next arrival, a trip plan or a service alert) goes back in the commuter's own language. The LLM never computes routes or times itself. We first review fifteen core studies on transit data, routing, arrival prediction, LLMs and Indic-language interfaces, and derive six design requirements from them. We then describe the system in full: objectives and scope, feasibility, an Agile development plan, data and functional analysis with data flow diagrams, hardware and software requirements, the layered architecture, the processing algorithm, the chat interface and the database schema. SafarSaathi needs no app install, works on low bandwidth, and can be extended to any city that publishes a GTFS feed. The paper closes with the evaluation protocol for the prototype and plans for future work.


Keywords

Public transport, Real-time passenger information, GTFS, OpenTripPlanner, Large language models, WhatsApp chatbot, Multilingual NLP, Digital inclusion



Full Text:

Download Paper PDF


References


Barnes, R., Buthpitiya, S., Cook, J., Fabrikant, A., Tomkins, A., & Xu, F. (2020). BusTr: Predicting bus travel times from real-time traffic. In Proceedings of the 26th ACM SIGKDD Conference on Knowledge Discovery and Data Mining. arXiv:2007.00882.

Bast, H., Delling, D., Goldberg, A., Müller-Hannemann, M., Pajor, T., Sanders, P., Wagner, D., & Werneck, R. F. (2016). Route planning in transportation networks. In Algorithm Engineering (LNCS Vol. 9220, pp. 19–80). Cham, Switzerland: Springer.

Brooke, J. (1996). SUS: A ‘quick and dirty’ usability scale. In P. W. Jordan, B. Thomas, B. A. Weerdmeester, & I. L. McClelland (Eds.), Usability evaluation in industry (pp. 189–194). London, U.K.: Taylor & Francis.

Brown, T., et al. (2020). Language models are few-shot learners. In Advances in Neural Information Processing Systems (pp. 1877–1901).

Delling, D., Pajor, T., & Werneck, R. F. (2015). Round-based public transit routing. Transportation Science, 49(3), 591–604.

Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2019). BERT: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of NAACL-HLT (pp. 4171–4186).

Devunuri, S., Qiam, S., & Lehe, L. J. (2024). ChatGPT for GTFS: Benchmarking LLMs on GTFS semantics... and retrieval. Public Transport, 16(2), 333–357.

Dibbelt, J., Pajor, T., Strasser, B., & Wagner, D. (2018). Connection scan algorithm. ACM Journal of Experimental Algorithmics, 23. arXiv:1703.05997.

Eltigani, H., Haroon, R., Kocak, A., Faisal, A. B., Martin, N., & Dogar, F. (2025). WaLLM – Insights from an LLM-powered chatbot deployment via WhatsApp. arXiv:2505.08894.

Fantin, L., Antonelli, M., Cesetti, M., Irto, D., Zamengo, B., & Silvestri, F. (2025). Design and testing of an agent chatbot supporting decision making with public transport data. arXiv:2505.22698.

Ferris, B., Watkins, K., & Borning, A. (2010). OneBusAway: Results from providing real-time arrival information for public transit. In Proceedings of ACM CHI (pp. 1807–1816).

Gala, J., et al. (2023). IndicTrans2: Towards high-quality and accessible machine translation models for all 22 scheduled Indian languages. Transactions on Machine Learning Research.

Gala, J., et al. (2024). Airavata: Introducing Hindi instruction-tuned LLM. arXiv:2401.15006.

Giannakopoulou, K., Paraskevopoulos, A., & Zaroliagis, C. (2018). Multimodal dynamic journey planning. arXiv:1804.05644.

Government of India. (2023). The Digital Personal Data Protection Act, 2023 (Act No. 22 of 2023).

He, M., & Garner, P. N. (2023). Can ChatGPT detect intent? Evaluating large language models for spoken language understanding. In Proceedings of Interspeech. arXiv:2305.13512.

Joshi, R. (2022). L3Cube-MahaCorpus and MahaBERT: Marathi monolingual corpus, Marathi BERT language models, and resources. arXiv:2202.01159.

Kakwani, D., et al. (2020). IndicNLPSuite: Monolingual corpora, evaluation benchmarks and pre-trained multilingual language models for Indian languages. In Findings of EMNLP (pp. 4948–4961).

Kitchenham, B. (2004). Procedures for performing systematic reviews (Technical Report TR/SE-0401). Keele, U.K.: Keele University.

Lewis, P., et al. (2020). Retrieval-augmented generation for knowledge-intensive NLP tasks. In Advances in Neural Information Processing Systems.

Liu, D., Guo, J., Gu, Y., King, M., Han, L. D., & Brakewood, C. (2024). GTFS2STN: Analyzing GTFS transit data by generating spatiotemporal transit network. arXiv:2405.02760.

Meta Platforms. (n.d.). WhatsApp Business Platform: Cloud API documentation. Retrieved from https://developers.facebook.com/docs/whatsapp/cloud-api/

MobilityData. (n.d.). General Transit Feed Specification: GTFS Schedule and GTFS Realtime reference. Retrieved from https://gtfs.org/

Narboneta, V., & Teknomo, K. (2016). OpenTripPlanner, OpenStreetMap, General Transit Feed Specification: Tools for disaster relief and recovery. arXiv:1609.01472.

OpenTripPlanner Project. (n.d.). OpenTripPlanner 2 documentation. Retrieved from https://docs.opentripplanner.org/

Page, M. J., et al. (2021). The PRISMA 2020 statement: An updated guideline for reporting systematic reviews. BMJ, 372, n71.

Radford, A., Kim, J. W., Xu, T., Brockman, G., McLeavey, C., & Sutskever, I. (2023). Robust speech recognition via large-scale weak supervision. In Proceedings of ICML.

Schick, T., et al. (2023). Toolformer: Language models can teach themselves to use tools. In Advances in Neural Information Processing Systems.

Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, Ł., & Polosukhin, I. (2017). Attention is all you need. In Advances in Neural Information Processing Systems (pp. 5998–6008).

Watkins, K. E., Ferris, B., Borning, A., Rutherford, G. S., & Layton, D. (2011). Where is my bus? Impact of mobile real-time information on the perceived and actual wait time of transit riders. Transportation Research Part A, 45(8), 839–848.

 

 

 

Share your valuable work from Social Media Buttons