Yara Rizk

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.

Yara Rizk, Praveen Venkateswaran, Vatche Isahagian et al. · ibm-research

The inception of large language models has helped advance state-of-the-art performance on numerous natural language tasks. This has also opened the door for the development of foundation models for other domains and data modalities such as images, code, and music. In this paper, we argue that business process data representations have unique characteristics that warrant the development of a new class of foundation models to handle tasks like process mining, optimization, and decision making. These models should also tackle the unique challenges of applying AI to business processes which include data scarcity, multi-modal representations, domain specific terminology, and privacy concerns.

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.

Siyu Huo, Kushal Mukherjee, Jayachandu Bandlamudi et al. · ibm-research

APIs are everywhere; they provide access to automation solutions that could help businesses automate some of their tasks. Unfortunately, they may not be accessible to the business users who need them but are not equipped with the necessary technical skills to leverage them. Wrapping these APIs with chatbot capabilities is one solution to make these automation solutions interactive. In this work, we propose a system to generate sentences to train intent recognition models, a crucial component within chatbots to understand natural language utterances from users. Evaluation of our approach based on deep learning models showed promising and inspiring results, and the human-in-the-loop interaction will provide further improvement on the system.

Ankita Bhaumik, Praveen Venkateswaran, Yara Rizk et al. · ibm-research

The popularity of conversational digital assistants has resulted in the availability of large amounts of conversational data which can be utilized for improved user experience and personalized response generation. Building these assistants using popular large language models like ChatGPT also require additional emphasis on prompt engineering and evaluation methods. Textual similarity metrics are a key ingredient for such analysis and evaluations. While many similarity metrics have been proposed in the literature, they have not proven effective for task-oriented conversations as they do not take advantage of unique conversational features. To address this gap, we present TaskDiff, a novel conversational similarity metric that utilizes different dialogue components (utterances, intents, and slots) and their distributions to compute similarity. Extensive experimental evaluation of TaskDiff on a benchmark dataset demonstrates its superior performance and improved robustness over other related approaches.

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.

Mayank Agarwal, Ibrahim Abdelaziz, Kinjal Basu et al.

As large language models (LLMs) increasingly interact with external tools, reward modeling for tool use has become a critical yet underexplored area. Existing reward models, trained primarily on natural language outputs, struggle to evaluate tool-based reasoning and execution. To quantify this gap, we introduce FC-RewardBench, the first benchmark designed to systematically assess reward models' performance in tool-calling scenarios. Our analysis shows that current reward models often miss key signals of effective tool use, highlighting the need for domain-specific modeling. To address this, we propose a training framework for outcome-based reward models using data synthesized from permissively licensed, open-weight LLMs. We train models ranging from 1.7B to 14B parameters and evaluate them across seven out-of-domain benchmarks. These models consistently outperform general-purpose baselines, achieving up to 25\% average improvement in downstream task performance and enabling data-efficient fine-tuning through reward-guided filtering.

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%.

Ibrahim Abdelaziz, Kinjal Basu, Mayank Agarwal et al.

Large language models (LLMs) have recently shown tremendous promise in serving as the backbone to agentic systems, as demonstrated by their performance in multi-faceted, challenging benchmarks like SWE-Bench and Agent-Bench. However, to realize the true potential of LLMs as autonomous agents, they must learn to identify, call, and interact with external tools and application program interfaces (APIs) to complete complex tasks. These tasks together are termed function calling. Endowing LLMs with function calling abilities leads to a myriad of advantages, such as access to current and domain-specific information in databases and knowledge sources, and the ability to outsource tasks that can be reliably performed by tools, e.g., a Python interpreter or calculator. While there has been significant progress in function calling with LLMs, there is still a dearth of open models that perform on par with proprietary LLMs like GPT, Claude, and Gemini. Therefore, in this work, we introduce the GRANITE-20B-FUNCTIONCALLING model under an Apache 2.0 license. The model is trained using a multi-task training approach on seven fundamental tasks encompassed in function calling, those being Nested Function Calling, Function Chaining, Parallel Functions, Function Name Detection, Parameter-Value Pair Detection, Next-Best Function, and Response Generation. We present a comprehensive evaluation on multiple out-of-domain datasets comparing GRANITE-20B-FUNCTIONCALLING to more than 15 other best proprietary and open models. GRANITE-20B-FUNCTIONCALLING provides the best performance among all open models on the Berkeley Function Calling Leaderboard and fourth overall. As a result of the diverse tasks and datasets used for training our model, we show that GRANITE-20B-FUNCTIONCALLING has better generalizability on multiple tasks in seven different evaluation datasets.

Burak Aksar, Yara Rizk, Tathagata Chakraborti · ibm-research

Chatbots have become one of the main pathways for the delivery of business automation tools. Multi-agent systems offer a framework for designing chatbots at scale, making it easier to support complex conversations that span across multiple domains as well as enabling developers to maintain and expand their capabilities incrementally over time. However, multi-agent systems complicate the natural language understanding (NLU) of user intents, especially when they rely on decentralized NLU models: some utterances (termed single intent) may invoke a single agent while others (termed multi-intent) may explicitly invoke multiple agents. Without correctly parsing multi-intent inputs, decentralized NLU approaches will not achieve high prediction accuracy. In this paper, we propose an efficient parsing and orchestration pipeline algorithm to service multi-intent utterances from the user in the context of a multi-agent system. Our proposed approach achieved comparable performance to competitive deep learning models on three different datasets while being up to 48 times faster.

Sohini Upadhyay, Vatche Isahagian, Vinod Muthusamy et al.

AI business process applications automate high-stakes business decisions where there is an increasing demand to justify or explain the rationale behind algorithmic decisions. Business process applications have ordering or constraints on tasks and feature values that cause lightweight, model-agnostic, existing explanation methods like LIME to fail. In response, we propose a local explanation framework extending LIME for explaining AI business process applications. Empirical evaluation of our extension underscores the advantage of our approach in the business process setting.

Sarath Sreedharan, Tathagata Chakraborti, Yara Rizk et al.

A new design of an AI assistant that has become increasingly popular is that of an "aggregated assistant" -- realized as an orchestrated composition of several individual skills or agents that can each perform atomic tasks. In this paper, we will talk about the role of planning in the automated composition of such assistants and explore how concepts in automated planning can help to establish transparency of the inner workings of the assistant to the end-user.

Tathagata Chakraborti, Vatche Isahagian, Rania Khalaf et al.

In this survey, we study how recent advances in machine intelligence are disrupting the world of business processes. Over the last decade, there has been steady progress towards the automation of business processes under the umbrella of ``robotic process automation'' (RPA). However, we are currently at an inflection point in this evolution, as a new paradigm called ``Intelligent Process Automation'' (IPA) emerges, bringing machine learning (ML) and artificial intelligence (AI) technologies to bear in order to improve business process outcomes. The purpose of this paper is to provide a survey of this emerging theme and identify key open research challenges at the intersection of AI and business processes. We hope that this emerging theme will spark engaging conversations at the RPA Forum.

Yara Rizk, Vatche Isahagian, Scott Boag et al.

Robotic process automation (RPA) has emerged as the leading approach to automate tasks in business processes. Moving away from back-end automation, RPA automated the mouse-click on user interfaces; this outside-in approach reduced the overhead of updating legacy software. However, its many shortcomings, namely its lack of accessibility to business users, have prevented its widespread adoption in highly regulated industries. In this work, we explore interactive automation in the form of a conversational digital assistant. It allows business users to interact with and customize their automation solutions through natural language. The framework, which creates such assistants, relies on a multi-agent orchestration model and conversational wrappers for autonomous agents including RPAs. We demonstrate the effectiveness of our proposed approach on a loan approval business process and a travel preapproval business process.

Yara Rizk, Vatche Isahagian, Merve Unuvar et al.

The ubiquity of smart phones and electronic devices has placed a wealth of information at the fingertips of consumers as well as creators of digital content. This has led to millions of notifications being issued each second from alerts about posted YouTube videos to tweets, emails and personal messages. Adding work related notifications and we can see how quickly the number of notifications increases. Not only does this cause reduced productivity and concentration but has also been shown to cause alert fatigue. This condition makes users desensitized to notifications, causing them to ignore or miss important alerts. Depending on what domain users work in, the cost of missing a notification can vary from a mere inconvenience to life and death. Therefore, in this work, we propose an alert and notification framework that intelligently issues, suppresses and aggregates notifications, based on event severity, user preferences, or schedules, to minimize the need for users to ignore, or snooze their notifications and potentially forget about addressing important ones. Our framework can be deployed as a backend service, but is better suited to be integrated into proactive conversational agents, a field receiving a lot of attention with the digital transformation era, email services, news services and others. However, the main challenge lies in developing the right machine learning algorithms that can learn models from a wide set of users while customizing these models to individual users' preferences.

Yara Rizk, Abhishek Bhandwalder, Scott Boag et al.

Business process automation is a booming multi-billion-dollar industry that promises to remove menial tasks from workers' plates -- through the introduction of autonomous agents -- and free up their time and brain power for more creative and engaging tasks. However, an essential component to the successful deployment of such autonomous agents is the ability of business users to monitor their performance and customize their execution. A simple and user-friendly interface with a low learning curve is necessary to increase the adoption of such agents in banking, insurance, retail and other domains. As a result, proactive chatbots will play a crucial role in the business automation space. Not only can they respond to users' queries and perform actions on their behalf but also initiate communication with the users to inform them of the system's behavior. This will provide business users a natural language interface to interact with, monitor and control autonomous agents. In this work, we present a multi-agent orchestration framework to develop such proactive chatbots by discussing the types of skills that can be composed into agents and how to orchestrate these agents. Two use cases on a travel preapproval business process and a loan application business process are adopted to qualitatively analyze the proposed framework based on four criteria: performance, coding overhead, scalability, and agent overlap.

Julia El Zini, Yara Rizk, Mariette Awad

Recurrent neural networks (RNN) have been successfully applied to various sequential decision-making tasks, natural language processing applications, and time-series predictions. Such networks are usually trained through back-propagation through time (BPTT) which is prohibitively expensive, especially when the length of the time dependencies and the number of hidden neurons increase. To reduce the training time, extreme learning machines (ELMs) have been recently applied to RNN training, reaching a 99\% speedup on some applications. Due to its non-iterative nature, ELM training, when parallelized, has the potential to reach higher speedups than BPTT. In this work, we present \opt, an optimized parallel RNN training algorithm based on ELM that takes advantage of the GPU shared memory and of parallel QR factorization algorithms to efficiently reach optimal solutions. The theoretical analysis of the proposed algorithm is presented on six RNN architectures, including LSTM and GRU, and its performance is empirically tested on ten time-series prediction applications. \opt~is shown to reach up to 845 times speedup over its sequential counterpart and to require up to 20x less time to train than parallel BPTT.