Data Mining and Modeling
The proliferation of machine learning means that learned classifiers lie at the core of many products across Google. However, questions in practice are rarely so clean as to just to use an out-of-the-box algorithm. A big challenge is in developing metrics, designing experimental methodologies, and modeling the space to create parsimonious representations that capture the fundamentals of the problem. These problems cut across Google’s products and services, from designing experiments for testing new auction algorithms to developing automated metrics to measure the quality of a road map.
Data mining lies at the heart of many of these questions, and the research done at Google is at the forefront of the field. Whether it is finding more efficient algorithms for working with massive data sets, developing privacy-preserving methods for classification, or designing new machine learning approaches, our group continues to push the boundary of what is possible.
Recent Publications
Fine-Grained Table Retrieval for Open-Domain Tabular Question Answering
Xingyu Ji
Wojciech Kosiuk
Madelon Hulsebos
Proceedings of the 11th Workshop on Automated Knowledge Base Construction (AKBC 2026), Association for Computational Linguistics
Preview abstract
This work introduces a fine-grained table retrieval framework for grounding large
language models in heterogeneous, open-domain relational data. Instead of encoding a query as a single vector, the approach decomposes natural
language queries into semantic components and embeds each independently, enabling
more precise matching of compositional query intent. These representations are used in a staged retrieval pipeline with component-level
search, connectivity-aware grouping, and reranking. Experiments on three TARGET benchmark corpora show consistent improvements in capped recall@k and stronger alignment between query intent and tabular structure over dense retrieval baselines, particularly for longer and more complex queries and when using lightweight embedding models.
View details
Preview abstract
Human-Computer Interaction research and design pedagogy rely on idealized process models, such as the Double Diamond, to describe how user experiences are designed. These models assume an orderly, linear design process that, while easy to understand, fails to capture the iterative and collaborative reality of professional practice. A few qualitative studies have successfully captured this complexity -- still, they often suffer from retrospective narrative smoothing and lack systemic scale. To understand how design unfolds in real products, we analyzed historical snapshots of 102 Figma files from a multi-national technology company and investigated the true trajectories of the design process at scale. Our analysis reveals that while the established process models might be applicable, the operational details are highly non-linear. Rather than a straight line from ideation toward completion, design advances are repeatedly reset to the ideation stage as feedback is received. We argue that by treating design files as operational telemetry, the industry can move beyond abstract frameworks to build practices and collaborative tools that support the non-linear realities of modern product development.
View details
Agent Gym: A Framework for Continuous Evaluation and Evolution of LLM Agents Through Human-in-the-Loop Feedback
Pouya Ghiasnezhad Omran
Michael Zimmermann
Duncan Cambridge
Ashmita Kapoor
Tanya Dixit
2026
Preview abstract
Large Language Model (LLM) agents deployed in production environments face a fundamental tension: the agent’s behavior is frozen at deployment time, while the business rules and edge cases it must handle continue to evolve. Existing approaches address agent construction and one-time evaluation but provide no structured mechanism for continuous post-deployment behavioral correction without modifying the agent’s source code. Most of the approaches offered in the market, require intense collection of logs and traces, and re-examining the agent design by the engineering team, a process which is heavy, long and negates the economical value of agentic transformation. We introduce Agent Gym, a modular, domain- agnostic framework that wraps any existing LLM-based agent in a continuous evaluation-and-evolution loop. The framework provides six composable capabilities — Act, Evaluate, Investigate, Correct, Learn, and Observe — organized across three architectural zones: a constitution layer that codifies domain knowledge in configuration artifacts, a runtime inference pipeline that chains acting, investigation, and adaptive correction, and a learning loop that enables subject matter experts to discover and validate new correction rules through natural language interaction. The key technical contributions include a hybrid deterministic-LLM correction engine with 21 condition operators and three-tier actions, a three-layer investigation architecture for ground-truth-free compliance validation, and a programmatic safety loop that guarantees rule correctness before human approval. We further introduce the Spec-to-Note Gap, an autoencoder-inspired view of agentic system transparency. An open-source reference implementation for invoice processing demonstrates that the framework is fully operational and ready for adoption.
View details
Preview abstract
Synthesizing unstructured research materials into manuscripts is an essential yet under-explored challenge in AI-driven scientific discovery. Existing autonomous writers are rigidly coupled to specific experimental pipelines, and produce superficial literature reviews. We introduce PaperOrchestra, a multi-agent framework for automated AI research paper writing. It flexibly transforms unconstrained pre-writing materials into publication-ready LaTeX manuscripts, including comprehensive literature synthesis and generated visuals, such as plots and conceptual diagrams. To evaluate performance, we present PaperWriting-Eval200 , the first standardized benchmark of reverse-engineered raw materials from 200 top-tier AI conference papers, alongside a suite of automated peer-review evaluators. PaperOrchestra significantly improves literature review depth and citation coverage, outperforming autonomous baselines with 90%–99% win rates and reaching human-level parity. It also outperforms autonomous baselines on overall manuscript quality by 14%–38%, as measured by side-by-side human evaluation.
View details
Autonomous Causal Inference Using Artificial Intelligence Agents
Teginder Singh
Justin Chen
Kiran Dattani
Sean Hamill
Jake Van Bochove
Technical Disclosure Commons (2025)
Preview abstract
Post-market drug-safety procedures, known as pharmacovigilance, require a team of skilled physicians or epidemiologists to manually investigate each adverse event. This disclosure describes artificial intelligence (AI) agent-based, cloud computing techniques that automate the end-to-end cognitive workflow of investigating potential adverse drug events (ADEs). Moving beyond current techniques, which only perform statistical signal detection, the described techniques enable the determination of biological causality. Upon receiving a statistical signal of an adverse event, plausible biological hypotheses that can explain the signal are autonomously generated. A set of specialized software agents are dispatched to forage for evidence across disparate data sources (real-world evidence platforms, scientific literature, genomic databases, etc.). The results generated by the agents are synthesized into findings that score the likelihood of a causal link. The final output is a detailed, auditable causality dossier that enables human safety experts to make faster, better-informed decisions.
View details
Preview abstract
This paper outlines a grammar of data analysis, as distinct from grammars of data manipulation. The primitives of this grammar are metrics and dimensions. We describe a Python implementation of this grammar called Meterstick, which is agnostic to the underlying data source, which may be a DataFrame or a SQL database.
View details
Some of our teams
×
![]()