News

A live demo of Samsara, the multimodal stream processing prototype introduced in our EDBT 2026 vision paper.

Program chair for the Ninth International Workshop on Exploiting Artificial Intelligence Techniques for Data Management, co-located with ACM SIGMOD/PODS 2026 in Bengaluru, India.

A workload-centric benchmark for studying the energy-aware trade-offs of LLMs in data-processing tasks, showing that energy efficiency and task effectiveness are not aligned.

I am building a research group on next-generation data systems that combine classical data system principles with modern AI techniques. Openings for PhD students and student assistants will be announced here.

A vision for streaming systems that treat multimodal LLMs as first-class operators, so queries can run in real time over text, images, audio and sensor data together.

A scalable framework for rapid traffic incident analysis over existing CCTV infrastructure, pairing real-time key-frame filtering with multimodal LLM causal reasoning.

Opening the Black-Box: Explaining Learned Cost Models for Databases brings explainability techniques to learned cost models, and Learned Cost Models for Query Optimization: From Batch to Streaming Systems carries them over to streaming workloads.

A study of learned cost models on the query optimization tasks they are meant to support, rather than on estimation accuracy alone.

Decentralized window aggregation for non-decomposable quantile functions, cutting the network overhead of centralized aggregation in IoT networks.

The programme supports early-career researchers in leading their own junior research group.

A benchmarking system for parallel and distributed stream processing, built to make performance results across systems and placements comparable.

ZEROTUNE learns zero-shot cost models for parallelism tuning in stream processing, and COSTREAM predicts operator placement costs in edge-cloud environments.