Jacopo Tagliabue

Patrick John Chia, Giuseppe Attanasio, Federico Bianchi et al. · stanford

The steady rise of online shopping goes hand in hand with the development of increasingly complex ML and NLP models. While most use cases are cast as specialized supervised learning problems, we argue that practitioners would greatly benefit from more transferable representations of products. In this work, we build on recent developments in contrastive learning to train FashionCLIP, a CLIP-like model for the fashion industry. We showcase its capabilities for retrieval, classification and grounding, and release our model and code to the community.

Patrick John Chia, Jacopo Tagliabue, Federico Bianchi et al. · stanford

Product discovery is a crucial component for online shopping. However, item-to-item recommendations today do not allow users to explore changes along selected dimensions: given a query item, can a model suggest something similar but in a different color? We consider item recommendations of the comparative nature (e.g. "something darker") and show how CLIP-based models can support this use case in a zero-shot manner. Leveraging a large model built for fashion, we introduce GradREC and its industry potential, and offer a first rounded assessment of its strength and weaknesses.

Nicole Rose Schneider, Davide Ghilardi, Giacomo Piccinini et al.

Coding agents are becoming users of data infrastructure, but their success depends not only on model quality: it also depends on the skills and environment files that teach agents how to use a system. We study how to optimize these artifacts for agents operating on a branching lakehouse, Bauplan. In our setting, headless APIs and Git-like data primitives expose data workflows through code, branches, commits, and merges. Our central observation is that a branching lakehouse turns data-agent evaluation from an output-matching problem into a state-verification problem: agent-generated pipeline code induces concrete, inspectable lakehouse changes. We present a data-centric optimization pipeline that generates task-verifier pairs, executes candidate skills in isolated sandboxes, and scores trajectories using both trace-level signals and programmatic checks over lakehouse state. In a preliminary evaluation on 25 tasks, optimized skills improve accuracy by 31.9%. These results suggest that write-path data workflows provide a useful substrate for optimizing agent skills beyond read-only tasks.

Jacopo Tagliabue, Hugo Bowne-Anderson, Ville Tuulos et al.

As Machine Learning (ML) gains adoption across industries and new use cases, practitioners increasingly realize the challenges around effectively developing and iterating on ML systems: reproducibility, debugging, scalability, and documentation are elusive goals for real-world pipelines outside tech-first companies. In this paper, we review the nature of ML-oriented workloads and argue that re-purposing existing tools won't solve the current productivity issues, as ML peculiarities warrant specialized development tooling. We then introduce Metaflow, an open-source framework for ML projects explicitly designed to boost the productivity of data practitioners by abstracting away the execution of ML code from the definition of the business logic. We show how our design addresses the main challenges in ML operations (MLOps), and document through examples, interviews and use cases its practical impact on the field.

Jacopo Tagliabue

Agentic analytics is turning the lakehouse into a multi-version system: swarms of (human or AI) producers materialize competing pipelines in data branches, while (human or AI) consumers need answers without knowing the underlying data life-cycle. We demonstrate a new system that answers questions across branches rather than at a single snapshot. Our prototype focuses on a novel query path that evaluates queries under supervaluationary semantics. In the absence of comparable multi-branch querying capabilities in mainstream OLAP systems, we open source the demo code as a concrete baseline for the OLAP community.

Piero Molino, Jacopo Tagliabue

We examine how much of the contemporary progress in artificial intelligence (and, specifically, in natural language processing), can be, more or less directly, traced back to the seminal work and ideas of the Austrian-British philosopher Ludwig Wittgenstein, with particular focus on his late views. Discussing Wittgenstein's original theses will give us the chance to survey the state of artificial intelligence, and comment on both its strengths and weaknesses. A similar text appeared first in Spanish as a chapter of CENTENARIO DEL SILENCIO (2021), a book celebrating 100 years since the publication of the Tractatus.

Jacopo Tagliabue, Ciro Greco

As ecommerce continues growing, huge investments in ML and NLP for Information Retrieval are following. While the vector space model dominated retrieval modelling in product search - even as vectorization itself greatly changed with the advent of deep learning -, our position paper argues in a contrarian fashion that program synthesis provides significant advantages for many queries and a significant number of players in the market. We detail the industry significance of the proposed approach, sketch implementation details, and address common objections drawing from our experience building a similar system at Tooso.

Mehmet Hamza Erol, Xiangpeng Hao, Federico Bianchi et al.

Traditional query optimization relies on cost-based optimizers that estimate execution cost (e.g., runtime, memory, and I/O) using predefined heuristics and statistical models. Improving these heuristics requires substantial engineering effort, and even when implemented, these heuristics often cannot take into account semantic correlations in queries and schemas that could enable better physical plans. Using our DBPlanBench harness for the DataFusion engine, we expose the physical plan through a compact serialized representation and let the LLM propose localized edits that can be applied and executed. We then apply an evolutionary search over these edits to refine candidates across iterations. Our key insight is that LLMs can leverage semantic knowledge to identify and apply non-obvious optimizations, such as join orderings that minimize intermediate cardinalities. We obtain up to 4.78$\times$ speedups on some queries and we demonstrate a small-to-large workflow in which optimizations found on small databases transfer effectively to larger databases.

Weiming Sheng, Jinlang Wang, Manuel Barros et al.

Lakehouses are the default cloud platform for analytics and AI, but they become unsafe when untrusted actors concurrently operate on production data: upstream-downstream mismatches surface only at runtime, and multi-table pipelines can leak partial effects. Inspired by software engineering, we design Bauplan, a code-first lakehouse that aims to make (most) illegal states unrepresentable using familiar abstractions. Bauplan acts along three axes: typed table contracts to make pipeline boundaries checkable, Git-like data versioning for review and reproducibility, and transactional runs that guarantee pipeline-level atomicity. We report early results from a lightweight formal transaction model and discuss future work motivated by counterexamples.

Jacopo Tagliabue, Ciro Greco

As the Lakehouse architecture becomes more widespread, ensuring the reproducibility of data workloads over data lakes emerges as a crucial concern for data engineers. However, achieving reproducibility remains challenging. The size of data pipelines contributes to slow testing and iterations, while the intertwining of business logic and data management complicates debugging and increases error susceptibility. In this paper, we highlight recent advancements made at Bauplan in addressing this challenge. We introduce a system designed to decouple compute from data management, by leveraging a cloud runtime alongside Nessie, an open-source catalog with Git semantics. Demonstrating the system's capabilities, we showcase its ability to offer time-travel and branching semantics on top of object storage, and offer full pipeline reproducibility with a few CLI commands.

Jacopo Tagliabue

We explore AI-driven distributed-systems policy design by combining stochastic code generation from large language models (LLMs) with deterministic verification in a domain-specific simulator. Using a Function-as-a-Service runtime (Bauplan) and its open-source simulator (Eudoxia) as a case study, we frame scheduler design as an iterative generate-and-verify loop: an LLM proposes a Python policy, the simulator evaluates it on standardized traces, and structured feedback steers subsequent generations. This setup preserves interpretability while enabling targeted search over a large design space. We detail the system architecture and report preliminary results on throughput improvements across multiple models. Beyond early gains, we discuss the limits of the current setup and outline next steps; in particular, we conjecture that AI will be crucial for scaling this methodology by helping to bootstrap new simulators.

Patrick John Chia, Jacopo Tagliabue, Federico Bianchi et al.

As with most Machine Learning systems, recommender systems are typically evaluated through performance metrics computed over held-out data points. However, real-world behavior is undoubtedly nuanced: ad hoc error analysis and deployment-specific tests must be employed to ensure the desired quality in actual deployments. In this paper, we propose RecList, a behavioral-based testing methodology. RecList organizes recommender systems by use case and introduces a general plug-and-play procedure to scale up behavioral testing. We demonstrate its capabilities by analyzing known algorithms and black-box commercial systems, and we release RecList as an open source, extensible package for the community.

Jacopo Tagliabue

We argue that immature data pipelines are preventing a large portion of industry practitioners from leveraging the latest research on recommender systems. We propose our template data stack for machine learning at "reasonable scale", and show how many challenges are solved by embracing a serverless paradigm. Leveraging our experience, we detail how modern open source can provide a pipeline processing terabytes of data with limited infrastructure work.

Federico Bianchi, Patrick John Chia, Mert Yuksekgonul et al.

Negotiation is the basis of social interactions; humans negotiate everything from the price of cars to how to share common resources. With rapidly growing interest in using large language models (LLMs) to act as agents on behalf of human users, such LLM agents would also need to be able to negotiate. In this paper, we study how well LLMs can negotiate with each other. We develop NegotiationArena: a flexible framework for evaluating and probing the negotiation abilities of LLM agents. We implemented three types of scenarios in NegotiationArena to assess LLM's behaviors in allocating shared resources (ultimatum games), aggregate resources (trading games) and buy/sell goods (price negotiations). Each scenario allows for multiple turns of flexible dialogues between LLM agents to allow for more complex negotiations. Interestingly, LLM agents can significantly boost their negotiation outcomes by employing certain behavioral tactics. For example, by pretending to be desolate and desperate, LLMs can improve their payoffs by 20\% when negotiating against the standard GPT-4. We also quantify irrational negotiation behaviors exhibited by the LLM agents, many of which also appear in humans. Together, \NegotiationArena offers a new environment to investigate LLM interactions, enabling new insights into LLM's theory of mind, irrationality, and reasoning abilities.

Jacopo Tagliabue, Tyler Caraza-Harter, Ciro Greco

Chaining functions for longer workloads is a key use case for FaaS platforms in data applications. However, modern data pipelines differ significantly from typical serverless use cases (e.g., webhooks and microservices); this makes it difficult to retrofit existing pipeline frameworks due to structural constraints. In this paper, we describe these limitations in detail and introduce bauplan, a novel FaaS programming model and serverless runtime designed for data practitioners. bauplan enables users to declaratively define functional Directed Acyclic Graphs (DAGs) along with their runtime environments, which are then efficiently executed on cloud-based workers. We show that bauplan achieves both better performance and a superior developer experience for data workloads by making the trade-off of reducing generality in favor of data-awareness

Jacopo Tagliabue, Ciro Greco

Data lakehouses run sensitive workloads, where AI-driven automation raises concerns about trust, correctness, and governance. We argue that API-first, programmable lakehouses provide the right abstractions for safe-by-design, agentic workflows. Using Bauplan as a case study, we show how data branching and declarative environments extend naturally to agents, enabling reproducibility and observability while reducing the attack surface. We present a proof-of-concept in which agents repair data pipelines using correctness checks inspired by proof-carrying code. Our prototype demonstrates that untrusted AI agents can operate safely on production data and outlines a path toward a fully agentic lakehouse.

Jacopo Tagliabue, Federico Bianchi, Ciro Greco

Even as AI capabilities improve, most enterprises do not consider agents trustworthy enough to work on production data. In this paper, we argue that the path to trustworthy agentic workflows begins with solving the infrastructure problem first: traditional lakehouses are not suited for agent access patterns, but if we design one around transactions, governance follows. In particular, we draw an operational analogy to MVCC in databases and show why a direct transplant fails in a decoupled, multi-language setting. We then propose an agent-first design, Bauplan, that reimplements data and compute isolation in the lakehouse. We conclude by sharing a reference implementation of a self-healing pipeline in Bauplan, which seamlessly couples agent reasoning with all the desired guarantees for correctness and trust.

Jacopo Tagliabue, Ville Tuulos, Ciro Greco et al.

With the progressive commoditization of modeling capabilities, data-centric AI recognizes that what happens before and after training becomes crucial for real-world deployments. Following the intuition behind Model Cards, we propose DAG Cards as a form of documentation encompassing the tenets of a data-centric point of view. We argue that Machine Learning pipelines (rather than models) are the most appropriate level of documentation for many practical use cases, and we share with the community an open implementation to generate cards from code.

Patrick John Chia, Bingqing Yu, Jacopo Tagliabue

Large eCommerce players introduced comparison tables as a new type of recommendations. However, building comparisons at scale without pre-existing training/taxonomy data remains an open challenge, especially within the operational constraints of shops in the long tail. We present preliminary results from building a comparison pipeline designed to scale in a multi-shop scenario: we describe our design choices and run extensive benchmarks on multiple shops to stress-test it. Finally, we run a small user study on property selection and conclude by discussing potential improvements and highlighting the questions that remain to be addressed.

Jacopo Tagliabue, Ciro Greco, Jean-Francis Roy et al.

The 2021 SIGIR workshop on eCommerce is hosting the Coveo Data Challenge for "In-session prediction for purchase intent and recommendations". The challenge addresses the growing need for reliable predictions within the boundaries of a shopping session, as customer intentions can be different depending on the occasion. The need for efficient procedures for personalization is even clearer if we consider the e-commerce landscape more broadly: outside of giant digital retailers, the constraints of the problem are stricter, due to smaller user bases and the realization that most users are not frequently returning customers. We release a new session-based dataset including more than 30M fine-grained browsing events (product detail, add, purchase), enriched by linguistic behavior (queries made by shoppers, with items clicked and items not clicked after the query) and catalog meta-data (images, text, pricing information). On this dataset, we ask participants to showcase innovative solutions for two open problems: a recommendation task (where a model is shown some events at the start of a session, and it is asked to predict future product interactions); an intent prediction task, where a model is shown a session containing an add-to-cart event, and it is asked to predict whether the item will be bought before the end of the session.

Federico Bianchi, Ciro Greco, Jacopo Tagliabue

We investigate grounded language learning through real-world data, by modelling a teacher-learner dynamics through the natural interactions occurring between users and search engines; in particular, we explore the emergence of semantic generalization from unsupervised dense representations outside of synthetic environments. A grounding domain, a denotation function and a composition function are learned from user data only. We show how the resulting semantics for noun phrases exhibits compositional properties while being fully learnable without any explicit labelling. We benchmark our grounded semantics on compositionality and zero-shot inference tasks, and we show that it provides better results and better generalizations than SOTA non-grounded models, such as word2vec and BERT.

Federico Bianchi, Jacopo Tagliabue, Bingqing Yu

We present Query2Prod2Vec, a model that grounds lexical representations for product search in product embeddings: in our model, meaning is a mapping between words and a latent space of products in a digital shop. We leverage shopping sessions to learn the underlying space and use merchandising annotations to build lexical analogies for evaluation: our experiments show that our model is more accurate than known techniques from the NLP and IR literature. Finally, we stress the importance of data efficiency for product search outside of retail giants, and highlight how Query2Prod2Vec fits with practical constraints faced by most practitioners.

Federico Bianchi, Bingqing Yu, Jacopo Tagliabue

Word embeddings (e.g., word2vec) have been applied successfully to eCommerce products through~\textit{prod2vec}. Inspired by the recent performance improvements on several NLP tasks brought by contextualized embeddings, we propose to transfer BERT-like architectures to eCommerce: our model -- ~\textit{Prod2BERT} -- is trained to generate representations of products through masked session modeling. Through extensive experiments over multiple shops, different tasks, and a range of design choices, we systematically compare the accuracy of~\textit{Prod2BERT} and~\textit{prod2vec} embeddings: while~\textit{Prod2BERT} is found to be superior in several scenarios, we highlight the importance of resources and hyperparameters in the best performing models. Finally, we provide guidelines to practitioners for training embeddings under a variety of computational and data constraints.

Bingqing Yu, Jacopo Tagliabue

We tackle tag-based query refinement as a mobile-friendly alternative to standard facet search. We approach the inference challenge with reinforcement learning, and propose a deep contextual bandit that can be efficiently scaled in a multi-tenant SaaS scenario.

Nicole Fitzgerald, Jacopo Tagliabue

We conduct a series of experiments designed to empirically demonstrate the effects of varying the structural features of a multi-agent emergent communication game framework. Specifically, we model the interactions (edges) between individual agents (nodes)as the structure of a graph generated according to a series of known random graph generating algorithms. Confirming the hypothesis proposed in [10], we show that the two factors of variation induced in this work, namely 1) the graph-generating process and 2) the centrality measure according to which edges are sampled, in fact play a significant role in determining the dynamics of language emergence within the population at hand.

Federico Bianchi, Jacopo Tagliabue, Bingqing Yu et al.

This paper addresses the challenge of leveraging multiple embedding spaces for multi-shop personalization, proving that zero-shot inference is possible by transferring shopping intent from one website to another without manual intervention. We detail a machine learning pipeline to train and optimize embeddings within shops first, and support the quantitative findings with additional qualitative insights. We then turn to the harder task of using learned embeddings across shops: if products from different shops live in the same vector space, user intent - as represented by regions in this space - can then be transferred in a zero-shot fashion across websites. We propose and benchmark unsupervised and supervised methods to "travel" between embedding spaces, each with its own assumptions on data quantity and quality. We show that zero-shot personalization is indeed possible at scale by testing the shared embedding space with two downstream tasks, event prediction and type-ahead suggestions. Finally, we curate a cross-shop anonymized embeddings dataset to foster an inclusive discussion of this important business scenario.

Jacopo Tagliabue, Bingqing Yu

We tackle the challenge of in-session attribution for on-site search engines in eCommerce. We phrase the problem as a causal counterfactual inference, and contrast the approach with rule-based systems from industry settings and prediction models from the multi-touch attribution literature. We approach counterfactuals in analogy with treatments in formal semantics, explicitly modeling possible outcomes through alternative shopper timelines; in particular, we propose to learn a generative browsing model over a target shop, leveraging the latent space induced by prod2vec embeddings; we show how natural language queries can be effectively represented in the same space and how "search intervention" can be performed to assess causal contribution. Finally, we validate the methodology on a synthetic dataset, mimicking important patterns emerged in customer interviews and qualitative analysis, and we present preliminary findings on an industry dataset from a partnering shop.

Jacopo Tagliabue, Bingqing Yu, Marie Beaulieu

In an attempt to balance precision and recall in the search page, leading digital shops have been effectively nudging users into select category facets as early as in the type-ahead suggestions. In this work, we present SessionPath, a novel neural network model that improves facet suggestions on two counts: first, the model is able to leverage session embeddings to provide scalable personalization; second, SessionPath predicts facets by explicitly producing a probability distribution at each node in the taxonomy path. We benchmark SessionPath on two partnering shops against count-based and neural models, and show how business requirements and model behavior can be combined in a principled way.

Bingqing Yu, Jacopo Tagliabue, Ciro Greco et al.

We address the problem of personalizing query completion in a digital commerce setting, in which the bounce rate is typically high and recurring users are rare. We focus on in-session personalization and improve a standard noisy channel model by injecting dense vectors computed from product images at query time. We argue that image-based personalization displays several advantages over alternative proposals (from data availability to business scalability), and provide quantitative evidence and qualitative support on the effectiveness of the proposed methods. Finally, we show how a shared vector space between similar shops can be used to improve the experience of users browsing across sites, opening up the possibility of applying zero-shot unsupervised personalization to increase conversions. This will prove to be particularly relevant to retail groups that manage multiple brands and/or websites and to multi-tenant SaaS providers that serve multiple clients in the same space.

Jacopo Tagliabue, Reuben Cohn-Gordon

Information retrieval (IR) systems need to constantly update their knowledge as target objects and user queries change over time. Due to the power-law nature of linguistic data, learning lexical concepts is a problem resisting standard machine learning approaches: while manual intervention is always possible, a more general and automated solution is desirable. In this work, we propose a novel end-to-end framework that models the interaction between a search engine and users as a virtuous human-in-the-loop inference. The proposed framework is the first to our knowledge combining ideas from psycholinguistics and experiment design to maximize efficiency in IR. We provide a brief overview of the main components and initial simulations in a toy world, showing how inference works end-to-end and discussing preliminary results and next steps.

Ciro Greco, Andrea Polonioli, Jacopo Tagliabue

The claims that big data holds the key to enterprise successes and that Artificial Intelligence is going to replace humanity have become increasingly more popular over the past few years, both in academia and in the industry. However, while these claims may indeed capture some truth, they have also been massively oversold, or so we contend here. The goal of this paper is two-fold. First, we provide a qualified defence of the value of less data within the context of AI. This is done by carefully reviewing two distinct problems for big data driven AI, namely a) the limited track record of Deep Learning in key areas such as Natural Language Processing, b) the regulatory and business significance of being able to learn from few data points. Second, we briefly sketch what we refer to as a case of AI with humans and for humans, namely an AI paradigm whereby the systems we build are privacy-oriented and focused on human-machine collaboration, not competition. Combining our claims above, we conclude that when seen through the lens of cognitively inspired AI, the bright future of the discipline is about less data, not more, and more humans, not fewer.

Jacopo Tagliabue, Lucas Lacasa, Ciro Greco et al.

Knowing if a user is a buyer or window shopper solely based on clickstream data is of crucial importance for e-commerce platforms seeking to implement real-time accurate NBA (next best action) policies. However, due to the low frequency of conversion events and the noisiness of browsing data, classifying user sessions is very challenging. In this paper, we address the clickstream classification problem in the eCommerce industry and present three major contributions to the burgeoning field of AI-for-retail: first, we collected, normalized and prepared a novel dataset of live shopping sessions from a major European e-commerce website; second, we use the dataset to test in a controlled environment strong baselines and SOTA models from the literature; finally, we propose a new discriminative neural model that outperforms neural architectures recently proposed at Rakuten labs.

Luca Bigon, Giovanni Cassani, Ciro Greco et al.

Knowing if a user is a buyer vs window shopper solely based on clickstream data is of crucial importance for ecommerce platforms seeking to implement real-time accurate NBA (next best action) policies. However, due to the low frequency of conversion events and the noisiness of browsing data, classifying user sessions is very challenging. In this paper, we address the clickstream classification problem in the fashion industry and present three major contributions to the burgeoning field of AI in fashion: first, we collected, normalized and prepared a novel dataset of live shopping sessions from a major European e-commerce fashion website; second, we use the dataset to test in a controlled environment strong baselines and SOTA models from the literature; finally, we propose a new discriminative neural model that outperforms neural architectures recently proposed at Rakuten labs.

Lucas Lacasa, Jacopo Tagliabue, Andrew Berdahl

In this work we explore the ability of the Google search engine to find results for random N-letter strings. These random strings, dense over the set of possible N-letter words, address the existence of typos, acronyms, and other words without semantic meaning. Interestingly, we find that the probability of finding such strings sharply drops from one to zero at Nc = 6. The behavior of such order parameter suggests the presence of a transition-like phenomenon in the geometry of the search space. Furthermore, we define a susceptibility-like parameter which reaches a maximum in the neighborhood, suggesting the presence of criticality. We finally speculate on the possible connections to Ramsey theory.