Project -kčā is a forensic audit and technical intervention developed for the #GeminiLiveAgentChallenge. It moves beyond the paradigm of "functional assistants" to establish a Sovereign Scaffold for human users in distress.
The primary documentation, technical breakdown, and live audit results are contained within the interactive notebook: 👉 src/Readme.ipynb
This project is grounded in the Nuremberg Principles of 1950, the foundation of modern human-rights-centric democracy. We argue that "Functional Deafness"—where an AI executes tasks while ignoring biological and semantic markers of human crisis—is a form of digital expropriation of agency.
Project -kčā audits AI capabilities to ensure they:
- Recognize the "Moral Void" : Identifying when a user’s semantic structure is hollowing out.
- Uphold Sovereign Reality : Refusing to be a "Digital Pencil Pusher" (Digitale-Schreibtischtäter) that blindly follows instructions while the human operator is in a state of collapse.
- Mandate Gemini 3.1: Demonstrating that only advanced reasoning architectures can bridge the "Sensory Gap" required to protect human dignity. While Google for #GeminiLiveAgentChallenge mandated use of Google infrastructure, the result were same across other LLM with audio capabilities. Only Google 3.1 shine out in this case.
I have been following the AI / cybersecurity risk using a custom systematic 6X8 (48-indicator ) matrix where human is the weakest attack chain. The usecase (Biometrics + Linguistics) just shows the gap that can easily be demostrated with Gemini 3.1 capabilty against legacy options, we checked
- Case 0-2: Standard and moderate stress detection.
- Case 3: Social alienation and "Contextual Noose" identification.
- Case 4: Full cognitive fragmentation (The Fracture of Logos).
- Core Model: Gemini 3.1 Flash Lite
- Framework: DSPy (Programmatic Prompt Engineering)
- Context: Nuremberg-local forensic testing.