StreetDesignAI: A Multi-Persona Evaluation System for Inclusive Infrastructure Design

TL;DR

StreetDesignAI introduces a persona-based, multi-agent evaluation framework for inclusive cycling infrastructure design. By grounding evaluation in street context and enabling parallel feedback from diverse cyclist personas, the system surfaces experiential conflicts and supports iterative, visualization-driven design adjustments. A within-subjects study with 26 professionals shows improvements in understanding diverse needs, confidence translating insights into design decisions, and intention to adopt the tool, compared to a chatbot baseline. The work demonstrates how conflict surfacing and parameter-level feedback can scaffold deliberate trade-off reasoning, with implications for broader AI-assisted infrastructure design beyond cycling lanes.

Abstract

Designing inclusive cycling infrastructure requires balancing competing needs of diverse user groups, yet designers often struggle to anticipate how different cyclists experience the same street. We investigate how persona-based multi-agent evaluation can support inclusive design by making experiential conflicts explicit. We present StreetDesignAI, an interactive system that enables designers to (1) ground evaluation in street context through imagery and map data, (2) receive parallel feedback from cyclist personas spanning confident to cautious users, and (3) iteratively modify designs while surfacing conflicts across perspectives. A within-subjects study with 26 transportation professionals demonstrates that structured multi-perspective feedback significantly improves designers' understanding of diverse user perspectives, ability to identify persona needs, and confidence in translating them into design decisions, with higher satisfaction and stronger intention for professional adoption. Qualitative findings reveal how conflict surfacing transforms design exploration from single-perspective optimization toward deliberate trade-off reasoning. We discuss implications for AI tools that scaffold inclusive design through disagreement as an interaction primitive.

Figures (9)

• Figure 1: System workflow of StreetDesignAI: The system consists of four main modules: (A) Evaluation Generation collects street-level imagery from Google Street View and road attributes from OpenStreetMap, then uses a fine-tuned GPT-4.1 model to generate safety ratings, comfort scores, and key factors for four cyclist personas and driver's perspective; (B) Design Adjustment allows users to configure design parameters (lane width, color, buffer type) and uses GPT-image-1 to render modified streetscapes, triggering re-evaluation of all personas; (C) Deep Analysis supports multi-turn dialogue for in-depth analysis of design preferences across different personas within a single scenario, with system responses prioritized by relevance; (D) Multi-Design Comparison includes D.1 Multi-Persona Discussion for comparing design preferences across multiple scenarios, and D.2 Design Comparison Visualization module for side-by-side rating comparison with highlighted differences.
• Figure 2: Study workflow: participants completed five phases: (1) pre-study survey on design confidence; (2-3) two design tasks using StreetDesignAI and ChatGPT in counterbalanced order, each with post-task surveys; (4) comparative reflection; and (5) semi-structured interview on system experiences and AI's role in inclusive design. Total session time averaged 95 minutes.
• Figure 3: Distribution of participant ratings for four key system functions (N=26). All functions rated above neutral midpoint.
• Figure 4: Frequency of design parameter selections across 48 design scenarios: (a) lane width, (b) lane color, (c) buffer type, (d) buffer location.
• Figure 5: Distribution of safety, comfort, and overall suitability scores across four cyclist personas (N=78 evaluations from 26 sessions).