CortexAI-kinematics

Inspiration

Agentic cinema projects often solve valuable production problems as isolated blocks. A real film production still needs a governed layer that can assemble evidence across domains, preserve human authority, record decisions, and translate approved direction into domain-specific objectives.

Cortex AI Kinematics addresses that integration gap. It is presented as a transparent, non-competitive integration perspective for Agentic Cinema: The Blockbuster Hackathon. The author participated by invitation, resides in Brazil, and does not claim prize eligibility.

What it does

Cortex AI Kinematics gives a production director one control plane for:

• opening governed deliberations;
• rotating evidence into working context;
• consulting agent adapters through the kernel;
• recording human decisions and audit events;
• decomposing approved direction into domain objectives;
• connecting autonomous cinema systems through a deliberately small SDK contract.

The Director Cockpit distinguishes verified local state, contract readiness, and missing infrastructure. It does not present planned capabilities as completed production operations.

Implemented and verified locally

• Kernel API and production session contracts
• Rotational context and governed working memory
• PostgreSQL deliberation and audit persistence
• Migration runner, health checks, and Prometheus metrics
• Director Cockpit
• Gemini executive-advisor adapter
• Cinema-domain connector SDK boundary
• Smoke tests and PostgreSQL integration tests

Connector delivery intentionally remains blocked until authenticated identity, authorization, a durable transactional outbox, connector registration, idempotency, signed acknowledgements, and immutable audit linkage exist end to end.

Architecture

The governing sequence is:

Director → identity and RBAC → Kernel API → governed deliberation → evidence and agent tools → human decision → durable transaction and outbox → connector runtime → production domain → signed acknowledgement → audit and cockpit.

Eligible production domains include Visual/VFX, Narrative, Finance, Safety, Audio, and Distribution. They remain autonomous behind the SDK contract.

Google Cloud production target

The adopter-owned production environment is designed for Cloud Run, Identity-Aware Proxy, Cloud SQL for PostgreSQL, Vertex AI Agent Engine with Gemini, Cloud Tasks, Pub/Sub, Secret Manager, Artifact Registry, Cloud Build, and Google Cloud observability.

The repository contains packaging foundations and architectural contracts, not a claim that this full environment is currently deployed. An organization adopting the system must provision and operate the required identity, infrastructure, connectors, reliability, security, and compliance controls.

A smaller agent-connected demonstration is planned separately.

How it was built

The production solution is implemented in C#/.NET with ASP.NET Core, PostgreSQL, a Gemini provider adapter, Docker packaging foundations, Prometheus metrics, Grafana artifacts, smoke tests, and an external connector SDK.

Challenges and learnings

The central challenge was not merely connecting agents, but defining an honest boundary between advice, human authority, durable state, and external execution. The project therefore keeps decision controls disabled until the infrastructure can prove delivery and acknowledgement.

AI-assisted development disclosure

Cortex AI Kinematics was developed with assistance from Gemini and OpenAI Codex. The owner estimates the contribution at approximately 50% from each as a good-faith process estimate, not a line-by-line authorship measurement. Gemini supported the initial implementation; Codex supported architectural review, production refactoring, testing, documentation, diagrams, and the executive presentation. Human responsibility remains with the project owner.

Participation note

This project does not claim prize eligibility or full judged-track compliance. It is a special, non-competitive presentation of an integration architecture intended to support participant solutions and potential winning systems during future development and adoption. Any exception involving eligibility, hosting, partner-track runtime integration, or submission format should remain subject to written organizer authorization.

What's next

• Connect a small demonstrative set of agents
• Add authenticated Director identity and RBAC
• Implement durable outbox/inbox delivery and connector acknowledgements
• Publish and version the SDK package
• Produce the public demonstration video
• Validate an adopter-owned Google Cloud deployment

Built With

• .net
• asp.net-core
• c#
• docker
• gemini-api
• google-cloud
• grafana
• postgresql
• prometheus