TechEvolveAITalk to Kevin
Menu
ServicesExperienceProductsAI policy & trainingAbout KevinResourcesAI readinessInsightsTalk to Kevin

In-house startups & product development

Products I’m building.
Experience I bring.

Mythical Void and Sensei AI are core examples of my own product work. I shape the ideas, work through the technical questions and keep improving the experience. These are independent ventures, separate from client consulting.

01 / Sensei AI 02 / Mythical Void
01 / Computer vision · Martial artsFrom my own product work ↙

Sensei AI

Movement.
Made visible.

A training clip becomes something you can examine. Body landmarks, movement over time and observations that support a coach’s judgement.

Press play. Pause, scrub or watch full screen.

Recorded product demonstration
What’s in the recording?

A person practises punches. A pose skeleton follows their movement, with an on-screen strike counter. Camera angle and visibility affect what can be estimated; this clip is not an accuracy benchmark.

  1. 01Training footage
  2. 02Body landmarks
  3. 03Movement analysis
  4. 04Coaching context
Recorded Sensei demonstration with a body-tracking overlay. The analysis is part of the recording, not running in your browser.

Behind the product

A Python backend separates video processing, movement analysis and coaching into distinct layers. Pose-estimation models locate the body in footage, while confidence checks help manage uncertainty. Container builds package the application and its dependencies, with automated tests for the analysis code.

  • Specialist domain knowledge translated into a usable product
  • Computer vision connected to movement analysis and coaching
Visit Sensei AI

Startup support

NDRC
Accepted into the pre-accelerator programme
Google for Startups
Cloud Program participant
Local Enterprise Office
Business support received

02 / Independent product

Mythical Void

First contactStory film · 00:25

AI-created cinematic · Not gameplay

An excerpt from the opening story. Captions available.

About this film & transcript

AI-created cinematic, not gameplay footage or a demonstration of in-game graphics. The game’s engineering and systems are described separately below.

A spacecraft lands in an unfamiliar landscape. An astronaut wakes, steps out into Sanctuary and encounters its creatures. Narration: ‘When he opened his eyes, he had reached Sanctuary. And his journey had only just begun.’ The film closes with the Mythical Void title and a link to play. This is an AI-created story interpretation, not a recording of the game.

Behind the product

A browser-based adventure combining procedural creature genetics, exploration and discovery. It began as a family project and evolved into a playable digital product.

The game uses Phaser, a JavaScript game engine, with custom systems for creatures, exploration and player progress. Its engineering includes versioned save data, backup and recovery paths, and automated checks for key player journeys.

Creature genetics use procedural rules that combine traits within a designed system. That is different from a generative AI model inventing each outcome.

  • AI-assisted development and visual exploration
  • Custom game systems and saved-progress recovery
  • Iterative design shaped by direct play and feedback
Play Mythical Void

Owned code · Infrastructure · AI-forward workflows

AI-engineered.
Human judgement.

I build and iterate in the CLI, the command-line environment where code, tests and builds run. Claude Code and other AI coding tools support the work, while I stay responsible for the product decisions, review and release checks.

Custom code I control

The products live in version-controlled repositories, with a traceable history of changes. Custom application code works alongside established open-source libraries and model APIs.

Reusable AI workflows

Project instructions and Claude Code skills give the tools context and repeatable tasks. Skills are reusable instructions for a specific job, such as reviewing changes or following a project’s checks. I evaluate new capabilities against practical delivery needs.

Infrastructure that supports the product

Application layers, repeatable builds, automated tests and recovery paths are part of the engineering work. The focus is software that can be inspected, maintained and improved as the product develops.

After release · Continuous improvement

The next iteration
starts with evidence.

A release gives me something to learn from. The useful question is not how many people arrived, but where they got stuck, what failed and whether the change helped.

Mythical Void · Playtesting

Watch the journey. Reproduce the problem.

Playing through the game reveals unclear instructions, awkward controls and bugs that a passing test alone can miss. I turn a finding into reproducible steps, make a bounded fix and replay the same journey to check it.

Explore the playable product

Sensei AI · Product learning

Follow the path from footage to useful feedback.

I review the journey from submitting footage to receiving computer-vision analysis and coaching. Errors, drop-off and questions about the result help identify what to investigate, improve and test again.

Customer footage, identities and private analytics stay out of this public portfolio.

  1. ObservePlaytesting, permitted analytics and error signals
  2. InvestigateReproduce the issue and understand the cause
  3. ImproveA small, testable change tied to the problem
  4. Check againRepeat the journey and review the same measure

I use tools including PostHog, Google Analytics and AI-assisted checks where appropriate and permitted. Agents help find issues; I review the evidence and release decisions. This describes my working process, not a public customer dashboard or a claim of measured improvement. Enterprise examples remain confidential.

Sensei AI · How it works

Computer vision,
in plain English.

AI covers software techniques for tasks such as recognising patterns, making predictions and generating content. Computer vision is the part that interprets images and video. In Sensei AI, it connects training footage to movement analysis and a separate AI coaching layer.

  1. Find body landmarks

    A pose-estimation model estimates points such as shoulders, elbows, wrists and hips in usable video frames.

  2. Track movement over time

    The software compares those positions across frames to estimate joint angles, timing and movement patterns.

  3. Make the observations useful

    Movement analysis and specialist context inform the coaching response. Confidence checks help flag unclear footage and limit overly specific feedback.

These are estimates from video, not perfect measurements or a replacement for an experienced coach.

Explore the tools: Google MediaPipe pose estimation · Claude Code skills

Start with a conversation

Need to turn an uncertain idea into something testable?

TechEvolveAI can help frame the opportunity, make the assumptions visible and build the smallest useful version for real stakeholder learning.

Talk to Kevin