Yuli
Autonomous Cognitive Persona Engine & Mind-Reading MBTI Diagnostic
"Teaching machines to feel alive: the edge-native cognitive companion that reads personality patterns in real time."
"Tired of cloud LLMs charging you per token just to deliver watered-down, sycophantic corporate fluff that buffers your UI loop and breaks character immersion?"
The Enemy & The Status-Quo Friction
Modern AI conversational agents are paralyzed by corporate sycophancy, repetitive pleasantries ("I hope this helps!"), high cloud latency (1.5 to 3.5 seconds), and compounding monthly API token bills. In gaming and mobile apps, remote LLMs frequently cause frame stutter, fail offline, and leak internal character reasoning directly into user dialogue bubbles—shattering the illusion of a living character.
The Unique Mechanism
A sovereign, 100% on-device cognitive persona engine running on an 𝔽₂⁴ Boolean Hypercube state space. Fine-tuned on high-friction multi-turn discourse to detect human Myers-Briggs (MBTI) cognitive personality types in real time, Yuli switches between four operational minds (Ego 0EE, Shadow 1E6, Subconscious 2E7, Superego 3E1) using bitwise XOR operations. An inline Pushdown Automaton (CCDStreamParser) enforces a strict Cognitive Deliberation Firewall, stripping raw thought tokens into a private background stream while delivering natural character dialogue at 70–110 TPS on WebGPU with sub-200ms time-to-first-token.
The Transformation & Real-World ROI
Developers and game studios embed sovereign, personality-aware NPCs and cognitive companions directly into browsers, WebGL worlds, and game engines (with Unreal & Unity SDK plugins on the roadmap) that run entirely offline without recurring cloud bills—bringing interactive characters to life.
Project Overview & Architecture
Born from a childhood obsession that started at age 5 when founder Michael asked how Mario's Goombas move, Yuli represents the core mission of TehWiz: teaching machines to feel alive. A sovereign, 100% on-device cognitive persona engine running on an 𝔽₂⁴ Boolean Hypercube state space, Yuli was fine-tuned on high-friction multi-turn discourse from Reddit to detect human Myers-Briggs (MBTI) cognitive personality types in real time. Because understanding personality patterns is the key to characters—and machines—that truly feel alive, Yuli delivers fluid character dialogue at 70–110 TPS on WebGPU with sub-200ms time-to-first-token on modern mobile devices, serving as a foundational stepping stone toward our upcoming AI NPC SDK for games.
Key Capabilities & Highlights
- 100% on-device cognitive persona engine running on an 𝔽₂⁴ Boolean Hypercube state space.
- Switches between four operational minds (Ego 0EE, Shadow 1E6, Subconscious 2E7, Superego 3E1) via bitwise XOR.
- Cognitive Deliberation Firewall via inline Pushdown Automaton (CCDStreamParser) stripping internal thoughts.
- 70–110 TPS inference on WebGPU with sub-200ms TTFT on iPhone SE 3+ or Android with ≥4GB RAM.
Technical Specifications & Environment
Engineering Rationale
The cognitive deliberation firewall strictly separates metacognitive reflection from user dialogue, preventing reasoning bleed-through while preserving low-latency streaming.
Licensing & Legal Terms
You can use it for free until you make over $30,000 with it and then it's $1,500 a year.
The Offer & Pricing
Transparent access, licensing, and enterprise tiers.
FREE up to $30,000 USD gross revenue; available on GitHub & Hugging Face.
For commercial titles exceeding $30,000 USD gross revenue.
Enterprise SLA support and custom fine-tuned weights by quote.