Keynote Speakers

Communities around the world are facing challenges due to their growing population and shrinking resources. The inter-disciplinary field of Smart City, also known as computational sustainability, has been exploring data-driven solutions for societal challenges for decades demonstrating many promising prototypes. Moreover, the past decade has been a great time for computing due to the rapid advances and availability in data, compute resources and analytical methods like Artificial Intelligence (AI) and Large Language Models (LLMs). However, the reality is that we are still far from tackling pressing societal problems involving human sustainability and wellbeing, including nutrition, exercises, collaboration, water, traffic. Why? In most cases, after promising initial prototypes, progress appears stalled in terms of realized improvement at scale and wider solution adoption. The barriers for this have been structural governance issues as well as technology limitations.

In this talk, we will focus on the technological aspects and how one may overcome persistent barriers. First, we will discuss a shift in perspective from one-off data-driven problem solving to incremental solution evolution based on increasing complexity considerations of data, methods and adoption based on trustworthy human-AI interaction. Then, we will use the specific instances of group recommendation for the AI method (that suggests groups, aka sets, bundles, of items to users based optimizing a combination of short and long term objectives over a time horizon), and robustness and fairness for trust. We will illustrate that more trustworthy, effective, solutions could be built via early case studies in collaboration (human-human teaming) and behavioral adherence (for nutrition in meals and physical activity) using domain-specific open, often big, data. The talk will also demonstrate the ULTRA team recommendation tool and the ARC tool for Blackbox AI assessment.

This talk delves into the principles, challenges, and opportunities involved in developing AI systems that are genuinely multilingual and culturally aware. As AI technologies increasingly influence global interactions, addressing the linguistic and cultural biases embedded in data, models, and evaluation frameworks becomes crucial. By drawing on examples from projects like Sanmati, Kahaani, and Pariksha, the talk underscores how participatory design and community-centric approaches can reveal nuanced biases and promote inclusive development. Additionally, it examines the role of synthetic datasets such as Updesh and benchmarking efforts like Samiksha in assessing AI across diverse languages and contexts. The session will highlight the significance of culturally contextual AI and the necessity of governance frameworks that reflect the lived realities of users in the Global South.

As Artificial Intelligence (AI) rapidly enters the education sector, critical questions emerge about its readiness, ethics, and alignment with pedagogical values. While many education systems embrace AI tools like ChatGPT or Gemini, such platforms often operate as opaque systems, raising concerns about algorithmic bias, misinformation, data privacy, and loss of teacher agency. In this context, the question “Are we ready?” becomes more relevant than ever.

This keynote explores how Kerala, through its public agency KITE (Kerala Infrastructure and Technology for Education), offers a pioneering and responsible model for AI in school education. With over 80,000 teachers trained in ethical AI usage and critical understanding, and AI introduced into the curriculum from Class 7 onward, the approach is rooted in pedagogy, equity, and transparency. The introduction of Robotics as a subject in Class 10 and the use of Free and Open-Source Software (FOSS) in over 15,000 schools further strengthen this model.

Kerala’s development of its own curriculum-aligned AI engine, Samagra Plus AI, using Retrieval-Augmented Generation (RAG) and curated local datasets, demonstrates how AI can serve public education goals. This keynote highlights the readiness Kerala has built—not just technologically, but ethically and pedagogically—to embrace AI in schools.

Edge accelerators have exponentially grown in their compute capabilities, forming a first-class computing fabric that extend to the cloud as well. This is of particular interest given the growing use of ML workloads for IoT data generated in the field. In this talk, we will examine the opportunities and challenges of modelling and optimizing edge accelerator platforms such as Nvidia Jetsons for ML workloads. It will explore our experiences with both single edge devices with heterogeneous accelerators, and distributed clusters of edge devices and cloud, to support diverse workloads from UAVs to federated learning. We will examine methods that help enhance performance metrics such as training throughput, inference latency and energy efficiency under constrained conditions. We will also discuss future research challenges of edge and cloud resources in the LLM era with agentic AI.