Rhetor: A Conversational Tool for Planning Under Scarcity
Rhetor: A Conversational Tool for Planning Under Scarcity
Many planning and allocation problems in public services, logistics, energy, agriculture, healthcare, education, and small industry can be expressed as mathematical programming models. In simple terms, mathematical programming is a way to describe a decision problem using variables, an objective, and practical constraints: for example, how to assign limited staff, distribute scarce supplies, schedule activities, or allocate budgets while satisfying operational rules. These methods are widely used in advanced planning, but access to suitable tools is often limited by software cost, licensing restrictions, and the technical effort required to build and maintain models.
Rhetor is a free integrated modelling environment designed for this setting. It supports the development and solution of mixed-integer linear programming models and offers three practical elements that are especially valuable for institutions with limited resources. First, it provides a free compiler for a subset of OPL-style syntax, allowing users to write models in a concise mathematical-programming language rather than building everything from scratch in low-level code. Second, it is connected to a free optimization stack based on the HiGHS solver, giving users access to a serious solver without requiring expensive commercial software. Third, it includes conversational modelling support intended to help the modeller formulate and refine models more efficiently.
This last point is important. The conversational component of Rhetor should not be understood as a promise to make everyone a modeller. Rather, it is meant to assist people who are already engaged in model development: lecturers, analysts, planners, engineers, consultants, researchers, students, and technically minded practitioners. In many real projects, a large share of effort is spent not on solving the model but on formulating it clearly, checking assumptions, stating variables and constraints correctly, and revising the structure as the real-world problem becomes better understood. Rhetor’s conversational support is valuable precisely because it can help speed up this modelling process. It can support drafting, clarifying, rephrasing, and iterating, while leaving judgement and responsibility with the modeller.
This makes Rhetor particularly relevant to developing and emerging countries. Many organizations in such settings face difficult optimization problems but cannot justify the cost of a full proprietary modelling environment. At the same time, they may still have staff or collaborators capable of formulating structured decision problems: for example, university departments, ministries, utilities, hospitals, NGOs, technical institutes, consultancies, innovation labs, and data teams. For these users, the main barrier is often not the absence of problems or the absence of talent, but the absence of an affordable and practical modelling stack, as well as the absence of time. Rhetor addresses that gap by combining a free modelling environment, an OPL-like language, a free solver backend, and workflow support that can reduce modelling time.
The practical applications are broad. A modeller might use Rhetor to develop a staff scheduling model for a hospital, an emergency relief dispatching plan, a school timetabling model, a district-level supply allocation model, a warehouse replenishment model, a transport assignment model, or a small production-planning model. In each case, the task is not merely to "run AI" on a vague problem description, but to develop a formal optimization model that reflects real operating constraints. Rhetor is valuable because it helps users carry out that development work more effectively and at lower cost.
For institutions in resource-constrained environments, this combination has several advantages. It supports teaching and training in mathematical programming using freely accessible tools. It enables pilots and operational prototypes without major licensing commitments. It can help build local capacity in model-based planning. And it may reduce dependence on expensive proprietary modelling ecosystems, while still offering a practical route to serious optimization.
Link to materials: Homepage: https://gwr3n.github.io/rhetor
Intro video: https://www.youtube.com/watch?v=o2Tm6ZByGc0
