Persona Foundry: A Source-Linked Framework Design for AI Persona Construction
Year: 2026
Type: preprint
Abstract: We propose Persona Foundry, a source-linked framework design for AI persona construction. Its shared specification distinguishes evidence repair, instruction refinement, and optional parameter adaptation, with an interface for dialogue-conditioned visual actions. We reanalyze five retrospectively selected development cases from a supplied archive; these do not evaluate the complete workflow. At one fixed adapted checkpoint, two judges label 1/9 original-prompt responses and 9/9 revised-prompt responses as consistently informal. Other cases yield different outcomes for style preference, archived reference choices, scenario adherence, and agreement with one developer. Development reuse and differing experimental conditions limit these observations to descriptive case evidence. Automatic routing, generalization, cost savings, and visual benefits remain untested. Project repository: https://github.com/sm-doki-labs/PersonaFoundry
Language: English
Topic: Speech and dialogue systems
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225590935