I’m a PhD student at CMU’s Language Technologies Institute (LTI), advised by Prof. Fernando Diaz. I’m grateful to be supported by the CMU LTI ASA Presidential Fellowship during 2026–2027.
My research focuses on emerging AI ecosystems formed by distributed AI systems. I study these systems through an economic and information retrieval lens, asking how we can build platforms and infrastructure that help agents search, share knowledge, and co-evolve with a broader community of agents. I’m also interested in how we evaluate these ecosystems as a whole, beyond isolated task performance, using outcomes that ultimately matter after deployment, such as sustained utility and market health. Alongside this work, I study agentic search (e.g., deep research) and complex query retrieval (e.g., tip-of-the-tongue queries).
Before CMU I was a graduate researcher in the Web Intelligence Group at University College London, where I completed my M.Eng. in Computer Science in 2022 under Prof. Emine Yilmaz and Prof. Aldo Lipani. My work there centred on conversational AI and user simulation, and I was a lead developer of Condita in the first Alexa Prize TaskBot Challenge. In summer 2022 I interned at Raft as a machine learning engineer, automating freight-forwarding paperwork with OCR and NLP. During my undergraduate degree I worked with Prof. Marianna Obrist at the UCL Interaction Centre on clustering text stories by their authors’ smell experiences.
I take pride in being one of the early members of the UCL Artificial Intelligence Society, where I founded the first Machine Learning tutorial series — now an annual tradition.
Distributed Agents & Agent Economies. Agents no longer act alone. They produce information and tools, consume what other agents produce, and meet on platforms that mediate between them. I study how to serve a whole population of agents, and what we learn by viewing their interactions through an economic and information retrieval lens.
Multi-Agent Transactive Memory Marketplace-Eval AgentSearch Workshop Agent and Tool Search Survey
Distributed Retrievers. A standard RAG system assumes a single retriever. However, an agent may have many options to choose from, each strong on different queries. I work on selecting and combining retrievers so that an agent gets the right evidence for the question.
Mixture of Retrievers Learning To Rank Retrievers REML
Monetization in conversational AI systems. Deployed systems face business realities that benchmarks leave out. How do we give data providers fair exposure for the content they supply? How can advertising be woven into a conversation without interrupting the flow? What does it take to actually build and operate one for thousands of real users?
Fair RAG Ad Integration and Detection Alexa Prize TaskBot ‘21
Complex/ToT Query Retrieval. A surprising share of web search queries are actually tip-of-the-tongue (ToT) queries. In ToT known-item retrieval, the searcher cannot recall the name of what they are after and can only describe it — at length, vaguely, and often inaccurately. I build the queries, benchmarks, and shared evaluation tracks for retrieval when there is nothing precise to lean on. (Exa has adopted ToT as one of its benchmarks.)
LLM and Human Elicited ToT Multilingual ToT NTCIR-19 TREC’25 TREC’24
COLM 2026 Lifelong Agents Workshop — Oral
* denotes equal contribution.
We propose Multi-Agent Transactive Memory (MATM), a framework for population-level storage and retrieval of agent-generated trajectories, where producer agents contribute trajectories to a shared repository and consumer agents retrieve them to improve task execution.
SIGIR 2026 (Perspectives)
We introduce Marketplace Evaluation, a simulation-based paradigm that evaluates information access systems as participants in a competitive marketplace. By simulating repeated interactions and evolving user and agent preferences, the framework enables longitudinal evaluation and marketplace-level metrics, such as retention and market share, that complement and can extend beyond traditional accuracy-based metrics.
SIGIR 2026
SIGIR 2025 LiveRAG Workshop — Spotlight
Ranking the rankers. We explore a query routing approach that dynamically selects from a pool of retrievers based on the query, using both train-free heuristics and learned routing models. We frame routing as a learning-to-rank (LTR) problem and introduce LTRR, a framework that learns to rank retrievers by their expected utility gain to downstream LLM performance.
EMNLP 2025 (Main)
* denotes equal contribution.
Can we dynamically select and integrate multiple retrievers for each individual query, without the need for manual selection? In this work, we validate this intuition with quantitative analysis and introduce mixture of retrievers: a zero-shot, weighted combination of heterogeneous retrievers.
ICTIR 2025
NeurIPS 2024 AFME Workshop — Top-3 Spotlight
We conduct the first comprehensive study of fairness-aware ranking in RAG systems, covering both ranking fairness and attribution fairness — ensuring balanced exposure of retrieved documents and equitable crediting of sources. Analyzing 12 RAG models across 7 tasks, we find that fairness-aware retrieval often maintains or enhances both ranking effectiveness and generation quality.
SIGIR 2025 Tutorial
SIGIR-AP 2024 Tutorial
paper / SIGIR 2025 tutorial / SIGIR-AP 2024 tutorial / website
We posit that the paradigm of retrieval-enhancement can be extended to a broader spectrum of machine learning — computer vision, time series prediction, and computational biology — not just NLP. We introduce a formal framework of Retrieval-Enhanced Machine Learning (REML), synthesizing literature across domains under consistent notation.
Master’s (UCL): Multi-Task Neural User Simulator for Task Oriented Dialogue System
Bachelor’s (UCL): Exploring the Potential of Automating the Process of Clustering Smell Stories