Noondra / Actor

Product context

Noondra is an independent platform for conversational agents. Actor, its principal product, explores roleplay agents that can maintain context, express a consistent character, speak, use constrained tools and appear through real-time 3D presentation.

The product challenge

A convincing interaction is not produced by an LLM alone. Context, memory, speech, tool permissions, presentation and recovery behaviour all influence whether an agent feels coherent and remains under product control. Each capability has its own failure modes, but the user experiences only one character.

End-to-end ownership

I design and implement the product across NestJS services, web and mobile clients, visual authoring tools, a Unity SDK and an on-device runtime. This gives me responsibility for both the architecture between surfaces and the interaction details inside them.

The platform boundary matters: Noondra carries reusable agent capabilities while Actor turns them into a focused product experience. That separation allows product decisions to guide the platform without making every capability dependent on a single interface.

A controllable local runtime

The local runtime composes LLM inference, conversation memory, speech, guardrails and constrained tool use. I treat those elements as collaborating subsystems with explicit responsibilities, not as a single opaque intelligence layer.

Keeping the runtime inspectable supports practical iteration: it becomes possible to reason about what context was available, which capability acted and where the experience broke down.

Designing for constraints

The central trade-off is between natural interaction and control. Local execution introduces real hardware limits; character consistency needs memory without uncontrolled accumulation; tools add usefulness but require constrained access. The design goal is not maximum capability in isolation, but a product that behaves coherently within known boundaries.

Current state

Noondra and Actor remain in active development. They represent independent full-stack product ownership and applied-AI systems engineering, while the quality and readiness of each experience continue to be evaluated rather than assumed.