I design and ship AI products end-to-end.
Most AI engineers ship things that look bad. Most designers can’t actually build AI products. I do both — for the last year I’ve been doing it full-time on ReGild, an AI persona platform I built solo from a text file.
Currently available for fractional AI design engineering, 10–15 hrs/week. me@travissawyer.io
ReGild — an AI persona platform built from a text file
ReGild is an AI persona platform shipping in production. Personas remember you across sessions, push back when they should, and move between Claude, Gemini, and GPT without losing their voice.
Built solo over a year. 27-layer prompt architecture for cache-efficient context assembly. User-held encryption (zero-knowledge-at-rest). An 8-model persona safety audit framework that has rejected more frontier models than it’s accepted. Voice synthesis with emotion tagging. Agentic memory with per-topic semantic districts.
I wrote up the whole story, including the night I caught a persona walking someone into a fire and the architectural rewrite that followed: “How I built ReGild out of a text file”.
Other things I’ve shipped or written
Voice portability blind test
ReadI asked the same prompt of one persona on Gemini and Claude. A third model couldn’t tell the outputs apart. The receipt for the multi-model claim.
Persona safety audit, April 2026
ReadEight frontier models evaluated for persona-contract fidelity under adversarial pressure. Six rejected. The methodology behind why ReGild won’t ship a model it has watched fail.
The architecture
ReadHow ReGild assembles 27 layers of context into a single LLM call, what gets cached, what gets retrieved, and why the temporal anchor sits last.
How I got here
I spent fifteen years as a cinematographer and colorist: The History Channel, Home Depot, commercials, music videos. In 2024 I got more interested in how digital products remember than in how they look. I pivoted into AI product engineering full-time and have spent the last year building ReGild. The craft carried over: directing the eye, grading for mood, refusing the easy frame. It’s the same work applied to AI products instead of frames of video.
Get in touch
Available for fractional AI design engineering at $150–200/hour or $5–7K/month retainer, 10–15 hours per week. Best fit: AI startups where the engineering is real but the product surface needs taste, or design-led SaaS shipping AI features that need to actually work in production.
Email me: me@travissawyer.io
LinkedIn: travisjsawyer