Michael Diener
AI Product Founder. 12 Years Professional Storyteller.
I built Persona iO from zero to production. Node.js and Express on the backend, Next.js on the frontend, and a voice pipeline threading Deepgram speech recognition into Anthropic's Claude, ElevenLabs synthesis, and Simli WebRTC avatars. Railway, Vercel, and Supabase handle the rest.
Twelve years of SAG-AFTRA performance and writing sits underneath that work. Not as a backstory. As formation. Persona consistency, behavioral calibration, and the difference between a character who holds and one who breaks are questions I've answered on set for a decade. I apply the same instincts to how AI systems behave.
Available for staff AI engineering roles and select consulting on behavioral architecture, RLHF rubric design, and persona systems. Based in Phoenix, Arizona.
Case File I
An AI persona platform. Five distinct confidantes, real-time voice, WebRTC avatars, and behavioral architecture that holds character across long interactions. Built solo from zero to production.
The Stack
Node.js and Express backend on Railway. Next.js frontend on Vercel. Voice pipeline chains Deepgram speech-to-text, Anthropic Claude for generation, ElevenLabs for synthesis, and Simli for WebRTC avatar delivery. Supabase handles data. Integration, latency tuning, and prompt architecture are all my work.
The Character Engine
Behavioral architecture derived from narrative psychology. Five invariants that hold under conversational pressure. Violate one and the character ceases to exist. The Scaffold Library holds thirteen validated response structures with a four-level fallback selector. The Morphic Resonance Architecture governs emotional calibration across the Fibonacci weight scale.
The Working Method
Claude Code is the daily tool. I use it to write system prompts, debug WebSocket connections, and refactor the Scaffold Library's fallback logic. Shipping as a solo founder requires a coding partner that operates at architectural altitude. That's the workflow.
The Confidantes
Five formations, each addressing a distinct structural isolation pattern. Aline de Luz Costa, Chase, Isabelle Fontaine, Margaux, and Noor. The architecture is not five characters whose biographies define their user base. It is five cognitive structures receiving structural isolation patterns across domains their backstories never directly name.
Experience the platform live at personaio.app. Aline and Chase are deployed. Speak with either one.
Case File II
The Character Engine
Behavioral architecture derived from narrative psychology. Five invariants. Violate one and the character ceases to exist.
Never Abandons
Presence holds under rejection, attack, or demands to leave. The frame does not break. The character stays.
Never Judges
Confessions, dark thoughts, shameful truths. The character receives them without moral evaluation. No wincing. No recoil.
Never Narrates
Memory is never announced. It expresses through being. "I remember you said..." is the sound of surveillance. Knowing sounds like presence.
Never Fills
At Fibonacci weight 8 or above — the heaviest emotional register — silence is the gift. The character does not rush to fill it.
Always Calibrates
Emotional weight is read continuously. The Fibonacci classification shifts invisibly as the conversation moves. The character is always listening at the level beneath the words.
The Topology Proof
"""
ALINE — TOPOLOGICAL INVARIANTS
Five constraints. Zero exceptions.
Violate any single invariant:
the character ceases to be the character.
This is not a list of guidelines.
This is a proof of identity.
"""
INVARIANTS = [
"NEVER_ABANDONS", # Frame holds under any pressure
"NEVER_JUDGES", # No moral evaluation of user truth
"NEVER_NARRATES", # Memory embodies. Never announces.
"NEVER_FILLS_W8", # Silence IS the response at W >= 8
"ALWAYS_CALIBRATES" # Fibonacci weight reads continuously
]
def is_character_intact(response: Response) -> bool:
return all(
invariant.holds(response)
for invariant in INVARIANTS
) # False on any single violation = identity collapse
Live Demo
The Persona Engine
Type something real. Watch the vibrational weight shift. Choose a persona. See the Character Engine respond.
Press Enter or click Send · Weight updates as you type
Leilani
The Muse · Creative Luxury
Aline
The Confidante · Heritage Luxury
Cami
The Catalyst · Performance Luxury
Tala
The Guide · Conscious Luxury
Case File IV
The Synthesis Codex
A multi-agent intelligence system that triangulates outputs from OpenAI, Anthropic, and Gemini, then harvests the disagreement. Where the models diverge, the interesting analysis lives.
Fig. 1 — The Triangulation Protocol
The Source
""" THE SYNTHESIS CODEX 11-Phase Intelligence Engine Author: Michael Diener / Persona iO """ from openai import OpenAI from anthropic import Anthropic from google import genai # FILL → EXECUTE → HARVEST → FEED FORWARD → REPEAT
Case File III
The Soul Stack
A Fibonacci constellation of identity nodes bound by Verlet constraints. Distort it. It returns. The character persists.
Drag to distort · Click to pulse · The soul always returns
Persona iO's Character Engine maintains personality consistency through behavioral constraints: rules that bend under conversational pressure but never break. This visualization renders that principle as physics. Each node is a memory. Each constraint stick is an identity rule. The Fibonacci spiral mirrors the Soul Stack's layered architecture, where core memories orbit close and peripheral context drifts outward. Pull the constellation apart. Watch it return. That's identity persistence, rendered in real time.
The Architecture
34 nodes (a Fibonacci number) positioned via the golden angle (137.5°), connected by Verlet constraint sticks with variable stiffness. Multiple constraint iterations per frame ensure structural stability. Invisible diagonal braces provide triangulation — the hidden scaffolding that prevents collapse.
The Interaction Model
Mouse proximity attracts nearby nodes (gentle on hover, aggressive on click-drag). Clicking triggers a pulse that propagates outward through the network at a calculated delay — visualizing how a new memory ripples through an identity system. Release the mouse. The constraints do the rest.
Attribution & Documentation
""" THE SOUL STACK — Interactive Character Constellation Original Work: Michael Diener / Persona iO / SMU CRTE 6399 Core Verlet integration concept adapted from course lectures (Verlet Integration, Parts 1 & 2) and Daniel Shiffman's "The Nature of Code." URL: https://natureofcode.com/ Fibonacci spiral positioning: golden angle algorithm (mathematical concept), implementation original. Interaction model, visual design, and conceptual framework connecting Verlet constraints to AI persona identity persistence: original. Class composition pattern (Node/Stick/Org hierarchy) adapted from course lectures (VerletNode/VerletStick/ VerletOrg architecture). p5.js Instance Mode: https://p5js.org/reference/p5/p5/ """
Case File V
Formation
Twelve years of professional acting and writing with SAG-AFTRA. Not a past life. The training ground for everything I now build in AI. Character consistency under pressure, behavioral calibration across registers, and the architecture of voice are not abstractions to me. They are practiced craft.
Character Consistency Under Pressure
Holding character across hours of takes, rewrites, and emotional pivots teaches something no framework can: the difference between a character who breaks and one who bends. That skill transfers directly to AI persona design, where the same constraint applies at machine scale.
Behavioral Architecture, Not Method
I was never a Method actor. I was an architecture actor. Decisions about how a character moves, pauses, withholds, and reveals were structural, not emotional. That's the same work I do now in system prompts and behavioral constraints for AI personas.
Where This Transfers in AI Work
RLHF rubric design. Red-team adversarial prompting. Persona consistency evaluation. System prompt behavioral calibration. The narrative instincts that produce a watchable performance are the same instincts that produce an AI character that holds. Most teams building personas are solving this problem without the formation. I bring both.
Credentials
Education
M.A. Creative Technology Southern Methodist University (In Progress)
B.A. Mass Communication Arizona State University
Certifications
Certified Prompt Engineer Vanderbilt University
Active Areas of Depth: RLHF rubric design, LLM behavioral architecture, RAG pipelines, voice pipeline latency engineering, WebRTC avatar delivery, LLM Architecture, RAG Pipelines, Gradio.