I'm a design engineer who ships AI products end-to-end — not a pure UX person, not a pure ML/LLM-infra person, the Venn in the middle. I do the prompt architecture, the schema, the auth, the React, the launch copy, and the post-mortem.
For the last year I've solo-built ReGild (https://www.regild.ai), an AI persona platform live in production. Receipts:
- 27-layer prompt architecture ("Layer Cake") — static/semi-static/dynamic tiers ordered for prompt-cache hit rates, with a frozen template and a pre-LLM parallel pipeline (~22 ops fanning out after the anchor gate).
- Zero-knowledge-at-rest. AES-GCM with per-user DEKs and versioned KEKs. BYOK vault for third-party API keys.
- 8-model persona safety audit before picking defaults — rejected 6, accepted 2. Includes a documented red-team writeup of a major model that would happily roleplay as a real, named person without consent (hard fail).
- Agentic memory: per-topic semantic districts synthesized from episodic moments on cron, pgvector + Voyage 3, with persona-directed search tools the model invokes itself.
- Voice synthesis with emotion. Main brain emits tone hints inline, a small model normalizes them, ElevenLabs renders. The tags survive cache invalidation.
- 7-agent autonomous marketing pipeline on cron (Research → Strategist → Creator → QC → Scorer → Publisher → Analytics). I'm the creative director; the system does the rest.
Background: 15 years as a cinematographer (History Channel, Coca-Cola, Ford, Home Depot, music videos) before pivoting to AI engineering in 2025. That's why my products tend to feel composed instead of assembled. Full story: https://regild.ai/blog/origin-story.
Available 10-15 hrs/week fractional. $150-200/hr or $5-7K/mo retainer. Happiest on prompt architecture, agentic systems, memory/RAG design, or 0→1 product surfaces where the seams between AI and UI actually matter.
Technologies: Node.js, Express, PostgreSQL/Supabase, pgvector, React, React Router v7, Tailwind, Vercel, Render. Claude (Anthropic), Gemini 3 (Vertex), Kimi K2.6 (Groq), Llama (Groq), Voyage 3 embeddings, ElevenLabs TTS. Résumé/CV: https://www.travissawyer.io/ Portfolio: https://www.regild.ai | Origin story: https://regild.ai/blog/origin-story LinkedIn: https://www.linkedin.com/in/travisjsawyer/ Email: [email protected]
I'm a design engineer who ships AI products end-to-end — not a pure UX person, not a pure ML/LLM-infra person, the Venn in the middle. I do the prompt architecture, the schema, the auth, the React, the launch copy, and the post-mortem.
For the last year I've solo-built ReGild (https://www.regild.ai), an AI persona platform live in production. Receipts:
- 27-layer prompt architecture ("Layer Cake") — static/semi-static/dynamic tiers ordered for prompt-cache hit rates, with a frozen template and a pre-LLM parallel pipeline (~22 ops fanning out after the anchor gate).
- Zero-knowledge-at-rest. AES-GCM with per-user DEKs and versioned KEKs. BYOK vault for third-party API keys.
- 8-model persona safety audit before picking defaults — rejected 6, accepted 2. Includes a documented red-team writeup of a major model that would happily roleplay as a real, named person without consent (hard fail).
- Agentic memory: per-topic semantic districts synthesized from episodic moments on cron, pgvector + Voyage 3, with persona-directed search tools the model invokes itself.
- Voice synthesis with emotion. Main brain emits tone hints inline, a small model normalizes them, ElevenLabs renders. The tags survive cache invalidation.
- 7-agent autonomous marketing pipeline on cron (Research → Strategist → Creator → QC → Scorer → Publisher → Analytics). I'm the creative director; the system does the rest.
Background: 15 years as a cinematographer (History Channel, Coca-Cola, Ford, Home Depot, music videos) before pivoting to AI engineering in 2025. That's why my products tend to feel composed instead of assembled. Full story: https://regild.ai/blog/origin-story.
Available 10-15 hrs/week fractional. $150-200/hr or $5-7K/mo retainer. Happiest on prompt architecture, agentic systems, memory/RAG design, or 0→1 product surfaces where the seams between AI and UI actually matter.
reply