Technology Showcases

Please note: Technology Showcase presentations are part of the conference Track Sessions. As such, they are educational (not commercial) and can feature
case studies which may include use of the exhibitor’s products and services. Technology Showcase content should also relate to the 7 Domains of the
INFORMS Analytics Framework.

Monday, April 13

10-10:50am

Description

10-10:50am

Bayesian Optimization: Iterative Learning for Optimizing Complex Products and Processes
Presented by: Ross Metusalem

Description
Bayesian Optimization (BayesOpt) is an iterative learning method for efficiently optimizing black-box or expensive-to-evaluate functions, with broad applications in process and product optimization. Starting with a small set of observations, BayesOpt fits a flexible machine learning-style model to the data and uses that model’s predictions and uncertainty in those predictions to determine the best next observation to collect. Iterating on this process, BayesOpt balances exploration the factor space with exploitation of potential optima to identify a globally optimal solution. This Technology Showcase will establish the foundational concepts underlying BayesOpt, discuss when to consider using it, and present a case study of the technique in action.

Related Domains:

Domain III: Data

Domain IV: Methodology (Approach) Framing

Domain V: Analytics/Model Development

Relevant to:

Essential (Early Career)

Professional (Mid-Career)

Executive (Senior Level)

11-11:50am

The Last Mile of OR: Building Production-Ready Decision Tools with DB Gene Studio
Presented by: Justin Clark

Description
You have a strong OR team. They’ve built sophisticated planning and scheduling models, tuned constraints, and found solutions that genuinely improve operations. But the models live in notebooks, the results get copy-pasted into spreadsheets, and the business stakeholders who need to act on the outputs can’t access them without going through a developer. The math is solved. The deployment isn’t.

This session presents a practical framework for closing that gap. Using a workforce planning problem as the working example, you’ll learn how to take a Python model all the way to a production-ready decision-support application — with interactive UIs business users can operate independently, AI-assisted constraint modeling using Claude, and seamless integration with existing data systems and solvers.

The development environment used throughout is DB Gene Studio, purpose-built for OR teams who want to own the full journey from model to deployment without dedicated engineering support. DB Gene Studio is solver-agnostic and designed to complement the tools and systems organizations already use.

Related Domain:

Domain V: Analytics/Model Development

Relevant to:

Professional (Mid-Career),

Executive (Senior Level) and

Associate (Early Career).

11-11:50am

GAMSPy: Powering Mathematical Optimization and Machine Learning in Python
Presented by: Adam Christensen & Steven Dirkse

Description
This showcase offers a glimpse into the world of modeling with GAMSPy – a powerful mathematical optimization package that merges the high-performance GAMS execution system with the flexibility of the Python language. The result is a complete mathematical optimization package that can be deployed anywhere – effortlessly.

Join us to explore GAMSPy’s fundamental functionalities through practical examples. We’ll cover everything from defining sets, parameters, variables, and equations to solving models and retrieving results, all within a familiar Python environment. Beyond the basics, we’ll also provide a glimpse into more advanced Machine Learning features, demonstrating how GAMSPy can streamline complex modeling workflows and enhance your analytical capabilities.

Whether you’re a seasoned GAMS user looking to integrate with Python or a Python user curious about optimization, this workshop will equip you with essential skills needed to get started and demonstrate what is possible with GAMSPy.

Related Domains:

Domain III: Data

Domain V: Analytics/Model Development

Domain VI: Deployment

Relevant to:

Essential (Early Career)

Professional (Mid-Career)

1-1:50pm

Overcoming the Limits of LLMs: Decision-Making Under Uncertainty with Causal Models
Presented by: Stefan Conrady

Descripton
Large language models (LLMs) have rapidly become central tools for analysis, explanation, and recommendation. Their apparent generality has led to a widespread perception that they can support virtually any analytics task. However, in real-world decision-making contexts, especially those involving uncertainty, interventions, and trade-offs, additional modeling capabilities are required.

This Technology Showcase introduces a practical framework for selecting and combining modeling approaches based on three key dimensions: learning granularity, representational abstraction, and inference semantics. Using this framework, we demonstrate how different analytics technologies, including statistical models, neural networks, LLMs, and Bayesian networks, serve distinct roles in the analytics workflow rather than acting as substitutes.

The session focuses on how Bayesian networks and influence diagrams enable explicit representation of uncertainty, causal reasoning, and decision optimization. Through live demonstrations and case examples, we show how these models support scenario analysis, intervention planning, and value-of-information calculations, helping organizations determine not only what is likely to happen, but what actions should be taken and what additional information is worth acquiring.

We also illustrate how LLMs can be integrated into this workflow to support knowledge elicitation, model structuring, and communication, while relying on structured probabilistic models for transparent and accountable decision support.

Attendees will leave with a practical understanding of how to position different modeling technologies within their analytics stack and how to build hybrid solutions that combine generative AI with causal and decision-theoretic modeling for high-stakes applications.

Related Domain:

Domain IV: Methodology

Relevant to:

Professional (Mid-Career)

1-1:50pm

The Smart Plant: AI-Powered Real-Time Optimization for Complex Industrial Operations
Presented by: Amit Garg and William Lopez

Description
In process-based industries, operational leadership faces a fundamental question: How can organizations continuously determine the optimal combination of raw materials to maximize product value, maintain asset health, control costs, and meet regulatory requirements—while operating in real time and at enterprise scale?

This presentation describes the architecture and implementation of an AI-driven Decision Intelligence platform developed by ORMAE that transforms this problem into a measurable source of operational advantage. The platform was initially designed to address copper concentrate blend optimization—one of the most computationally intensive problems in base metal smelting. Its design principles and technology architecture are broadly applicable to industrial operations where feed composition, procurement decisions, and process stability are tightly interconnected.

The system integrates combinatorial optimization, physics-based metallurgy and process simulation, AI agent orchestration, and enterprise system integration. It evaluates millions of possible input combinations within minutes and dynamically responds to changing operating conditions, supply disruptions, and commercial constraints. High-performance computing enables large-scale sourcing scenario analysis, while integration with plant control systems creates a closed-loop link between digital recommendations and physical operations.

Deployment results demonstrate significant improvements across smelting operations. Decision time for concentrate blending was reduced from 72 hours to less than 6 hours, procurement efficiency improved substantially, plant stability increased, and cross-functional collaboration strengthened across procurement, operations, finance, and commercial teams.

Attendees from metals, refining, chemicals, cement, fertilizers, battery materials, and advanced manufacturing will gain a replicable framework for implementing AI-driven decision intelligence in complex industrial environments.

Related Domains:

Domain I: Business Problem (Question) Framing

Domain II: Analytics Problem Framing

Domain III: Data

Domain IV: Methodology (Approach) Framing

Domain V: Analytics/Model Development

Domain VI: Deployment

Domain VII: Analytics Solution Lifecycle Management

Relevant to:

Essential (Early Career)

Professional (Mid-Career)

Executive (Senior Level)

2-2:50pm

An AI You Can Trust: Driving Analytics Adoption in Complex Manufacturing through Acceptability
Presented by: Louis-Pierre Campeau

Description
In the production of highly customized goods with extensive production chains, traditional predictive models face the “black box” barrier. While technical accuracy is often the focus of analytics developments, the ultimate success of an initiative strongly depends on user acceptability and integration into daily decision-making.

This presentation details the development and implementation of a predictive model designed for a complex manufacturing environment. We shift the focus from pure algorithmic performance to a methodology rooted in collaborative design and transparency. By involving end users throughout the development lifecycle, we ensured the tool was not only technically robust but also intuitive and trustworthy for those managing the actual work.

We will explore the specific methodological framework used to bridge the gap between data science and operational reality. Along the way, we will investigate the concept of acceptability, its implication for the success of a project and common biases. This case study demonstrates how non-technical aspects of a technical project can truly be the driving force of adoption.

Related Domains:

Domain I: Business Problem (Question) Framing

Domain II: Analytics Problem Framing

Domain III: Data

Domain IV: Methodology (Approach) Framing

Domain V: Analytics/Model Development

Relevant to:

Essential (Early Career)

Professional (Mid-Career)

Executive (Senior Level)

2-2:50pm

Best Practices for Decision Model Management: Versioning, Rollout, CI/CD, DecisionOps, and More
Presented by: Ryan O’Neil

Description
What input was used to run the model? What model version did it run on? Where are we tracking new model development? How long will it take to reproduce the experiment setup that generated the most recent test results? Is it safe to roll out the newest model iteration? And do we have a fallback model we can leverage just in case?

All of these questions are part of everyday decision model development. Answering them efficiently is critical to innovating within the discipline and driving sustained project success within teams and organizations. Too often teams rely on fragile or incomplete infrastructure to support these workflows, if they’re lucky to have them at all. Harness the DecisionOps habits that leading teams have used to drive greater and more sustained project success.

Join this session to learn about best practices for managing the full decision model lifecycle: from versioning, environment setup, CI/CD integration, model drift monitoring, rollout strategies, and more.

Related Domains:

Domain V – Analytics/Model Development

Domain VI – Deployment

Domain VII – Analytics Solution Lifecycle Management

Relevant to:

Essential (Early Career),

Professional (Mid-Career), and

Executive (Senior Level).

4-4:50pm

Optimization in Action: Decision Systems Across Industries
Presented by: Christian Valente, Jurgen Lentz, and Juan Bohorquez

Description
Optimization plays a critical role in operational decision systems across industries such as energy, transportation, logistics, retail, and finance. These systems combine predictive analytics, data pipelines, and mathematical optimization models to support decisions involving scheduling, resource allocation, routing, and planning.

This technology showcase highlights how optimization is used in real-world applications across a range of industries. We begin by illustrating where optimization fits within modern decision architectures and the role it plays as the optimization layer connecting data, models, and operational systems.

The session then explores several industry applications that use optimization to support complex operational decisions. Examples will demonstrate how organizations structure optimization models, integrate them into analytics environments, and deploy them within production systems.

Throughout the presentation, we highlight features and modeling approaches that enable scalable decision applications and explain why optimization continues to be a core technology for solving large operational problems across industries.

This talk is intended for operations researchers, data scientists, and analytics practitioners interested in how optimization is applied in modern decision systems.

Related Domains:

Domain I: Business Problem (Question) Framing

Domain V – Analytics/Model Development

Domain VI – Deployment

Relevant to:

Essential (Early Career),

Professional (Mid-Career), and

Executive (Senior Level).

4-4:50pm

Stop Optimizing Point Forecasts: Robust Decision Modeling with PyMC and FICO® Xpress
Presented by: Daniel Saunders, Evan Wimpey, and Jay Laramore

Description
Optimization models often rely on point forecasts that ignore uncertainty. Yet many operational decisions must remain robust against unpredictable demand, weather, and market conditions.

This talk presents a practical framework for building optimization models in the face of uncertainty by combining Bayesian probabilistic forecasting with mathematical optimization. Using PyMC, we generate posterior demand distributions that capture seasonality, trends, and extreme events. These posterior samples form a distribution of possible outcomes that are incorporated directly into optimization models that are solved using FICO Xpress.

We’ll demonstrate how this approach enables risk-aware decision-making using chance-constrained optimization and Conditional Value-at-Risk (CVaR). Through an electricity generation planning example, attendees will learn how probabilistic forecasts can be incorporated into FICO Xpress to support practical decision-making despite uncertainty.

Related Domains:

Domain IV

Domain V

Relevant to:

Essential (Early Career)

Professional (Mid-Career)

Tuesday, April 14

9:30-10:20am

Nonlinear Programming with Hexaly
Presented by: Fred Gardi

Description
Hexaly 14.5 is a hybrid optimization solver that combines exact and heuristic methods in a model-and-run framework. This work presents its approaches for continuous nonlinear problems min f(x) s.t. g_i(x) ≤ 0, h_j(x) = 0, bounds on x, with analytical functions modeled via an expression graph. New features in 14.5 version include: automatic structure detection and convexity classification by propagating properties through the graph to select suitable algorithms; a modular interior-point solver offering predictor-corrector, trust-region, and filter methods (chosen dynamically, with parallel multi-solve runs), falling back to derivative-free search or augmented Lagrangian when Hessians are unavailable; and an exact MINLP solver that uses model reformulation, RLT cuts, branch-and-reduce for tight bounds, and subproblem solves for primal solutions. In 60-second benchmarks, Hexaly 14.5 finds more feasible solutions and proves optimality more often than SCIP 9.2 and Ipopt 3.14 on convex QP portfolio instances, nonconvex QCQP pooling, and CUTEst nonlinear problems.

Related to:

Domain V: Analytics/Model Development

Relevant to:

Essential (Early Career)

Professional (Mid-Career)

Executive (Senior Level)

9:30-10:20am

Optimization Solution Development and Deployment – A Framework for Success
Presented by: Irv Lustig

Description

For over 44 years, Princeton Consultants has developed and deployed numerous applications that leverage optimization to make better decisions for our clients. We have developed a set of best practices over that time that are also reflected in the recently published INFORMS Analytics Framework (IAF), for which Irv was a key contributor.

The IAF provides a roadmap for successful analytics projects, including optimization, and contains specific tasks related to risk management. Irv will describe both the IAF and the Princeton 20, a set of 10 environmental and 10 technical risks to be evaluated at the commencement of any project. Understanding these risks in the context of the framework leads to best practices that improve the chances of success for your next optimization project.

11-11:50am

Selecting Appropriate Simulation User Interfaces
Presented by: Saurabh Parakh and Martin Franklin

Description
A simulation user interface often takes a model from code to a tool from which organizations gain value. This presentation focuses on why companies should invest in user interfaces for simulation models, what the different types of user interfaces are, and how an appropriate user interface option can be selected.

When selecting the appropriate user interface, several dimensions should be considered, including the number of users, user sophistication, data feeds to and from other applications, the frequency of model use, and the significance of the decision being made with the simulation model. This presentation will share examples of real-world projects with a variety of user interfaces to discuss why they were selected for their particular project.

At a minimum, a simulation user interface, sometimes referred to as the Front End, provides the user with a means to change model inputs and view model outputs. Additional functionality may include scenario management or optimization to increase the usefulness of a simulation model.

While there may be several reasons to create a user interface for a model, there are 3 significant reasons justifying the time required to design a good user interface:

It saves time and time is money. If users can reach the same decisions faster, it allows for a lower long-term cost of ownership for using the simulation model.

It prevents mistakes. If users are making the wrong decisions from a simulation model, then the model has done more harm than good.

It increases model adoption. If a model has the power to enable profit-optimal decisions, but no one wants to use it because it is too difficult, then the model is not useful to the organization.

Related Domains:

Domain I: Business Problem (Question) Framing

Domain IV: Methodology (Approach) Framing

Domain VI: Deployment

Relevant to:

Professional (Mid-Career)

Executive (Senior Level)

11-11:50am

Building and Solving Optimization Models with SAS
Presented by: Rob Pratt

Description
SAS offers extensive analytic capabilities, including machine learning, deep learning, natural language processing, statistical analysis, optimization, and simulation. SAS analytic functionality is also available through the open, cloud-enabled design of SAS® Viya®. You can program in SAS or in other languages – Python, Lua, Java, and R.

OPTMODEL from SAS provides a powerful and intuitive algebraic optimization modeling language and unified support for building and solving LP, MILP, QP, conic, NLP, constraint programming, network-oriented, and black-box models. This showcase will include an overview of the optimization capabilities and demonstrate recently added features.

Related Domain:

Domain V: Analytics/Model Development

Relevant to:

Essential (Early Career)

Professional (Mid-Career)

Executive (Senior Level)

Technology Showcases

Monday, April 13

Tuesday, April 14

Get the Latest Updates