HABTIC

I was COO, responsible for operations, production architecture, and the organizational structure that turned an impossible brief into shipping technology.

The challenge

HABTIC’s stakeholders had a clear, uncompromising vision:

• A health and wellbeing coaching app
• A cast of realistic, human-like avatars
• Full offline functionality
• Continuous, adaptive learning from user behavior

None of these requirements were possible with existing technology. Conventional "digital human" solutions relied on cloud streaming and heavy infrastructure, while game-style avatars required terabytes of pre-baked content. Meanwhile, AI models were too resource-intensive and too slow to adapt dynamically on-device.

In short: the project demanded technology that did not yet exist.

Core problems

• Avatar realism: no existing engine could support hours of interactive, personalized dialogue and animation offline.
• Offline constraint: streaming-based solutions were incompatible with the requirement for fully local operation.
• Adaptive intelligence: continuous learning and path adjustment could not rely on backend inference cycles or retraining.
• Device limitations: hundreds of hours of content and behavior needed to run efficiently on consumer mobile hardware.

Solution approach

HABTIC became a technology-creation program. Two core innovations were developed to solve the unsolvable:

TALOS: the autonomous avatar engine

A bespoke animation and behavior system that:

• uses looping motion primitives recombined dynamically by layered AI.
• creates the illusion of fully responsive, lifelike coaches without streaming or massive asset payloads.
• generates real-time reactions and behaviors tailored to user context.

Result: A responsive, expressive digital coach engine that operates entirely offline. To this day, completely unmatched in its class.

SYBIL: the adaptive pathfinding engine

A novel AI architecture inspired by flocking behavior and dynamic decision-making, designed to:

• choose the next best action based on real-time context, without needing fixed start/end states.
• continuously adapt to evolving user data and content libraries.
• operate without constant retraining or expensive backend compute.

Result: A unique, real-time prognostication system that remains state-aware, lightweight, and scalable.

What this taught me

• Hard, uncompromising stakeholder requirements drove the team past what the market thought was possible. Clarity of intent is an underrated catalyst.
• Every constraint the team encountered, including ones considered impossible, was solved by reframing the problem and engineering from first principles.
• The best outcomes came from listening deeply to domain experts and investing in experimentation before committing to architecture.

Do not fear stakeholders who don’t know what they want — fear those who do. They will force you to build the future.