Kiran Kate

Kinjal Basu, Ibrahim Abdelaziz, Kiran Kate et al. · ibm-research

The resurgence of autonomous agents built using large language models (LLMs) to solve complex real-world tasks has brought increased focus on LLMs' fundamental ability of tool or function calling. At the core of these agents, an LLM must plan, execute, and respond using external tools, APIs, and custom functions. Research on tool calling has gathered momentum, but evaluation benchmarks and datasets representing the complexity of the tasks have lagged behind. In this work, we focus on one such complexity, nested sequencing, with the goal of extending existing benchmarks and evaluation. Specifically, we present NESTFUL, a benchmark to evaluate LLMs on nested sequences of API calls, i.e., sequences where the output of one API call is passed as input to a subsequent call. NESTFUL contains 1800+ nested sequences where all the function calls are executable. Experimental results on a variety of models show that the best-performing model (GPT-4o) achieves a full sequence match accuracy of 28% and a win-rate of 60%, necessitating a large scope for improvement in the nested sequencing aspect of function calling. Our analysis of these results provides possible future research directions for the community, in addition to a benchmark to track progress. We have released the NESTFUL dataset under the Apache 2.0 license at https://github.com/IBM/NESTFUL.

Michael Feffer, Martin Hirzel, Samuel C. Hoffman et al.

Bias mitigators can improve algorithmic fairness in machine learning models, but their effect on fairness is often not stable across data splits. A popular approach to train more stable models is ensemble learning, but unfortunately, it is unclear how to combine ensembles with mitigators to best navigate trade-offs between fairness and predictive performance. To that end, we built an open-source library enabling the modular composition of 8 mitigators, 4 ensembles, and their corresponding hyperparameters, and we empirically explored the space of configurations on 13 datasets. We distilled our insights from this exploration in the form of a guidance diagram for practitioners that we demonstrate is robust and reproducible.

Zidane Wright, Jason Tsay, Anupama Murthi et al.

As AI agents move from demos into enterprise deployments, their failure modes become consequential: a misinterpreted tool argument can corrupt production data, a silent reasoning error can go undetected until damage is done, and outputs that violate organizational policy can create legal or compliance risk. Yet, most agent frameworks leave builders to handle these failure modes ad hoc, resulting in brittle, one-off safeguards that are hard to reuse or maintain. We present the Agent Lifecycle Toolkit (ALTK), an open-source collection of modular middleware components that systematically address these gaps across the full agent lifecycle. Across the agent lifecycle, we identify opportunities to intervene and improve, namely, post-user-request, pre-LLM prompt conditioning, post-LLM output processing, pre-tool validation, post-tool result checking, and pre-response assembly. ALTK provides modular middleware that detects, repairs, and mitigates common failure modes. It offers consistent interfaces that fit naturally into existing pipelines. It is compatible with low-code and no-code tools such as the ContextForge MCP Gateway and Langflow. Finally, it significantly reduces the effort of building reliable, production-grade agents.

Krishnateja Killamsetty, Alexandre V. Evfimievski, Tejaswini Pedapati et al.

Training deep networks and tuning hyperparameters on large datasets is computationally intensive. One of the primary research directions for efficient training is to reduce training costs by selecting well-generalizable subsets of training data. Compared to simple adaptive random subset selection baselines, existing intelligent subset selection approaches are not competitive due to the time-consuming subset selection step, which involves computing model-dependent gradients and feature embeddings and applies greedy maximization of submodular objectives. Our key insight is that removing the reliance on downstream model parameters enables subset selection as a pre-processing step and enables one to train multiple models at no additional cost. In this work, we propose MILO, a model-agnostic subset selection framework that decouples the subset selection from model training while enabling superior model convergence and performance by using an easy-to-hard curriculum. Our empirical results indicate that MILO can train models $3\times - 10 \times$ faster and tune hyperparameters $20\times - 75 \times$ faster than full-dataset training or tuning without compromising performance.

Karl Munson, Anish Savla, Chih-Kai Ting et al.

Style is a significant component of natural language text, reflecting a change in the tone of text while keeping the underlying information the same. Even though programming languages have strict syntax rules, they also have style. Code can be written with the same functionality but using different language features. However, programming style is difficult to quantify, and thus as part of this work, we define style attributes, specifically for Python. To build a definition of style, we utilized hierarchical clustering to capture a style definition without needing to specify transformations. In addition to defining style, we explore the capability of a pre-trained code language model to capture information about code style. To do this, we fine-tuned pre-trained code-language models and evaluated their performance in code style transfer tasks.

Michael Feffer, Martin Hirzel, Samuel C. Hoffman et al.

There are several bias mitigators that can reduce algorithmic bias in machine learning models but, unfortunately, the effect of mitigators on fairness is often not stable when measured across different data splits. A popular approach to train more stable models is ensemble learning. Ensembles, such as bagging, boosting, voting, or stacking, have been successful at making predictive performance more stable. One might therefore ask whether we can combine the advantages of bias mitigators and ensembles? To explore this question, we first need bias mitigators and ensembles to work together. We built an open-source library enabling the modular composition of 10 mitigators, 4 ensembles, and their corresponding hyperparameters. Based on this library, we empirically explored the space of combinations on 13 datasets, including datasets commonly used in fairness literature plus datasets newly curated by our library. Furthermore, we distilled the results into a guidance diagram for practitioners. We hope this paper will contribute towards improving stability in bias mitigation.

Lilian Ngweta, Kiran Kate, Jason Tsay et al.

Large language models (LLMs) have gained popularity in recent years for their utility in various applications. However, they are sensitive to non-semantic changes in prompt formats, where small changes in the prompt format can lead to significant performance fluctuations. In the literature, this problem is commonly referred to as prompt brittleness. Previous research on prompt engineering has focused mainly on developing techniques for identifying the optimal prompt for specific tasks. Some studies have also explored the issue of prompt brittleness and proposed methods to quantify performance variations; however, no simple solution has been found to address this challenge. We propose Mixture of Formats (MOF), a simple and efficient technique for addressing prompt brittleness in LLMs by diversifying the styles used in the prompt few-shot examples. MOF was inspired by computer vision techniques that utilize diverse style datasets to prevent models from associating specific styles with the target variable. Empirical results show that our proposed technique reduces style-induced prompt brittleness in various LLMs while also enhancing overall performance across prompt variations and different datasets.

Shubham Gandhi, Jason Tsay, Jatin Ganhotra et al.

Large Language Model (LLM) agents are increasingly deployed for complex, multi-step software engineering (SWE) tasks. However, their trajectories often contain costly inefficiencies, such as redundant exploration, looping, and failure to terminate once a solution is reached. Prior work has largely treated these errors in a post-hoc manner, diagnosing failures only after execution. In this paper, we introduce SWE-PRM, an inference-time Process Reward Model (PRM) that intervenes during execution to detect and course-correct trajectory-level errors. Our PRM design leverages a taxonomy of common inefficiencies and delivers lightweight, interpretable feedback without modifying the underlying policy. On SWE-bench Verified, closed-source PRMs improve resolution from 40.0% to 50.6% (+10.6 p.p.), with the largest gains on medium and hard tasks. Among feedback strategies, taxonomy-guided PRMs outperform unguided or explicit action-prescriptive variants, increasing success rate while reducing trajectory length. These benefits come at an acceptable added inference cost of as low as $0.2, making PRMs a practical and scalable mechanism for improving SWE agents' reliability and efficiency.

Benjamin Elder, Anupama Murthi, Jungkoo Kang et al.

Large language models (LLMs) are routinely deployed as agentic systems, with access to tools that interact with live environments to accomplish tasks. In enterprise deployments these systems need to interact with API collections that can be extremely large and complex, often backed by databases. In order to create datasets with such characteristics, we explore how existing NL2SQL (Natural Language to SQL query) datasets can be used to automatically create NL2API datasets. Specifically, this work describes a novel data generation pipeline that exploits the syntax of SQL queries to construct a functionally equivalent sequence of API calls. We apply this pipeline to one of the largest NL2SQL datasets, BIRD-SQL to create a collection of over 2500 APIs that can be served as invocable tools or REST-endpoints. We pair natural language queries from BIRD-SQL to ground-truth API sequences based on this API pool. We use this collection to study the performance of 10 public LLMs and find that all models struggle to determine the right set of tools (consisting of tasks of intent detection, sequencing with nested function calls, and slot-filling). We find that models have extremely low task completion rates (7-47 percent - depending on the dataset) which marginally improves to 50 percent when models are employed as ReACT agents that interact with the live API environment. The best task completion rates are far below what may be required for effective general-use tool-calling agents, suggesting substantial scope for improvement in current state-of-the-art tool-calling LLMs. We also conduct detailed ablation studies, such as assessing the impact of the number of tools available as well as the impact of tool and slot-name obfuscation. We compare the performance of models on the original SQL generation tasks and find that current models are sometimes able to exploit SQL better than APIs.

Jason Tsay, Zidane Wright, Gaodan Fang et al.

Agentic systems interact with external systems by calling tools such as Python functions, REST API endpoints, or command line tools such as kubectl in Kubernetes. These tool calls often fail for various syntactic and semantic reasons. Some less obvious semantic errors can only be identified and resolved after analyzing the tool's response. To repair these errors, we develop a post-tool execution reflection component that combines large language model (LLM)-based reflection with domain-specific retrieval-augmented generation (RAG) using documents describing both the specific tool being called and troubleshooting documents related to the tool. For this paper, we focus on the use case of the kubectl command line tool to manage Kubernetes, a platform for orchestrating cluster applications. Through a larger empirical study and a smaller manual evaluation, we find that our RAG-based reflection will repair kubectl commands such that they are both more likely to successfully execute (pass rate) for 55% of our models evaluated and 36% more likely to correctly answer the user query on average. We find that troubleshooting documents improve pass rate compared to official documentation by an average of 10%.

Koren Lazar, Matan Vetzler, Kiran Kate et al.

AI agents and business automation tools interacting with external web services require standardized, machine-readable information about their APIs in the form of API specifications. However, the information about APIs available online is often presented as unstructured, free-form HTML documentation, requiring external users to spend significant time manually converting it into a structured format. To address this, we introduce OASBuilder, a novel framework that transforms long and diverse API documentation pages into consistent, machine-readable API specifications. This is achieved through a carefully crafted pipeline that integrates large language models and rule-based algorithms which are guided by domain knowledge of the structure of documentation webpages. Our experiments demonstrate that OASBuilder generalizes well across hundreds of APIs, and produces valid OpenAPI specifications that encapsulate most of the information from the original documentation. OASBuilder has been successfully implemented in an enterprise environment, saving thousands of hours of manual effort and making hundreds of complex enterprise APIs accessible as tools for LLMs.

Karl Munson, Chih-Kai Ting, Serenity Wade et al.

Like text, programs have styles, and certain programming styles are more desirable than others for program readability, maintainability, and performance. Code style transfer, however, is difficult to automate except for trivial style guidelines such as limits on line length. Inspired by the success of using language models for text style transfer, we investigate if code language models can perform code style transfer. Code style transfer, unlike text transfer, has rigorous requirements: the system needs to identify lines of code to change, change them correctly, and leave the rest of the program untouched. We designed CSB (Code Style Benchmark), a benchmark suite of code style transfer tasks across five categories including converting for-loops to list comprehensions, eliminating duplication in code, adding decorators to methods, etc. We then used these tests to see if large pre-trained code language models or fine-tuned models perform style transfer correctly, based on rigorous metrics to test that the transfer did occur, and the code still passes functional tests. Surprisingly, language models failed to perform all of the tasks, suggesting that they perform poorly on tasks that require code understanding. We will make available the large-scale corpora to help the community build better code models.

Nikitha Rao, Jason Tsay, Kiran Kate et al.

Low-code programming allows citizen developers to create programs with minimal coding effort, typically via visual (e.g. drag-and-drop) interfaces. In parallel, recent AI-powered tools such as Copilot and ChatGPT generate programs from natural language instructions. We argue that these modalities are complementary: tools like ChatGPT greatly reduce the need to memorize large APIs but still require their users to read (and modify) programs, whereas visual tools abstract away most or all programming but struggle to provide easy access to large APIs. At their intersection, we propose LowCoder, the first low-code tool for developing AI pipelines that supports both a visual programming interface (LowCoder_VP) and an AI-powered natural language interface (LowCoder_NL). We leverage this tool to provide some of the first insights into whether and how these two modalities help programmers by conducting a user study. We task 20 developers with varying levels of AI expertise with implementing four ML pipelines using LowCoder, replacing the LowCoder_NL component with a simple keyword search in half the tasks. Overall, we find that LowCoder is especially useful for (i) Discoverability: using LowCoder_NL, participants discovered new operators in 75% of the tasks, compared to just 32.5% and 27.5% using web search or scrolling through options respectively in the keyword-search condition, and (ii) Iterative Composition: 82.5% of tasks were successfully completed and many initial pipelines were further successfully improved. Qualitative analysis shows that AI helps users discover how to implement constructs when they know what to do, but still fails to support novices when they lack clarity on what they want to accomplish. Overall, our work highlights the benefits of combining the power of AI with low-code programming.

Grady Booch, Francesco Fabiano, Lior Horesh et al.

This paper proposes a research direction to advance AI which draws inspiration from cognitive theories of human decision making. The premise is that if we gain insights about the causes of some human capabilities that are still lacking in AI (for instance, adaptability, generalizability, common sense, and causal reasoning), we may obtain similar capabilities in an AI system by embedding these causal components. We hope that the high-level description of our vision included in this paper, as well as the several research questions that we propose to consider, can stimulate the AI research community to define, try and evaluate new methodologies, frameworks, and evaluation metrics, in the spirit of achieving a better understanding of both human and machine intelligence.

Guillaume Baudart, Martin Hirzel, Kiran Kate et al.

Automated machine learning makes it easier for data scientists to develop pipelines by searching over possible choices for hyperparameters, algorithms, and even pipeline topologies. Unfortunately, the syntax for automated machine learning tools is inconsistent with manual machine learning, with each other, and with error checks. Furthermore, few tools support advanced features such as topology search or higher-order operators. This paper introduces Lale, a library of high-level Python interfaces that simplifies and unifies automated machine learning in a consistent way.

Guillaume Baudart, Peter D. Kirchner, Martin Hirzel et al.

AI automation tools need machine-readable hyperparameter schemas to define their search spaces. At the same time, AI libraries often come with good human-readable documentation. While such documentation contains most of the necessary information, it is unfortunately not ready to consume by tools. This paper describes how to automatically mine Python docstrings in AI libraries to extract JSON Schemas for their hyperparameters. We evaluate our approach on 119 transformers and estimators from three different libraries and find that it is effective at extracting machine-readable schemas. Our vision is to reduce the burden to manually create and maintain such schemas for AI automation tools and broaden the reach of automation to larger libraries and richer schemas.

Martin Hirzel, Kiran Kate, Avraham Shinnar et al.

Machine-learning automation tools, ranging from humble grid-search to hyperopt, auto-sklearn, and TPOT, help explore large search spaces of possible pipelines. Unfortunately, each of these tools has a different syntax for specifying its search space, leading to lack of portability, missed relevant points, and spurious points that are inconsistent with error checks and documentation of the searchable base components. This paper proposes using types (such as enum, float, or dictionary) both for checking the correctness of, and for automatically searching over, hyperparameters and pipeline configurations. Using types for both of these purposes guarantees consistency. We present Lale, an embedded language that resembles scikit learn but provides better automation, correctness checks, and portability. Lale extends the reach of existing automation tools across data modalities (tables, text, images, time-series) and programming languages (Python, Java, R). Thus, data scientists can leverage automation while remaining in control of their work.

Subhrajit Roy, Kiran Kate, Martin Hirzel

Systems that can automatically analyze EEG signals can aid neurologists by reducing heavy workload and delays. However, such systems need to be first trained using a labeled dataset. While large corpuses of EEG data exist, a fraction of them are labeled. Hand-labeling data increases workload for the very neurologists we try to aid. This paper proposes a semi-supervised learning workflow that can not only extract meaningful information from large unlabeled EEG datasets but also make predictions with minimal supervision, using labeled datasets as small as 5 examples.