Starting point01
The pieces are there.
A thumbnail, title, channel name, and multiplier.
Four years of compounding practice
I build the systems around intelligence.
I have worked deeply with generative AI since 2022—not as a layer added at the end, but as a new material for products, teams, and operating systems.
My work spans productized personalized feedback, adaptive roleplay, AI-readable design systems, agent teams, production delivery, and evaluation tools.
4years building with generative AI
End to endproduct strategy, design, and working software
Personal + teamindependent experiments and production tools for teams
00 The thesis
This is not a prompt practice. It is a systems practice.
The model is only one part of the work. Dependable AI needs structured context, bounded jobs, observable behavior, explicit quality standards, and feedback that improves the next run.
01Product systemsExperiences people use
02Operating modelsSystems teams can scale
03R&D infrastructureTools for what comes next
01 / Design systems + craftThe quality system
See what the system
makes possible.
A credible starting point. Then room to go further.
01 / A shared visual languageTubeBuddy · Content Outliers
Crafted further02
Outlier videos
39:27
SmarterEveryDay1.1M viewsvs 262K avg4.2x
3.8K comments920 views/hr
12M subs3mo ago
Refined component
They feel like one product.
Shared components give identity and performance a consistent voice.
Look closer
The same avatar and multiplier components carry across cards and details.
Select a detail to trace it in the cards.More people can build.
Shared components and product context give the team a credible starting point.
I can take quality further.
AI-assisted implementation lets me refine hierarchy, behavior, and the details directly in working software.
Adapted from TubeBuddy component history · illustrative matched metrics
Behind this comparison
These interactive examples adapt the original product components for this portfolio. The imagery is real; the matched metrics and request timing are illustrative. Highlight outlines are added for inspection.
The card and playback refinements came from the August 2026 Outliers work. The Silk Sweep loading state and filter-change feedback followed in September. Playback startup here is a simulation, not a live video or a readiness event.
The motion preview exposes a shared app flag. The product honors the device’s reduced-motion preference; a persistent user-facing product switch is not established by the inspected source. The remaining pulse is intentional in this historical version.
I refined the design in code; engineering retained approval and release ownership. Read the production-system case ↗
Foundation test5 distinct product experimentsCompared component foundations against real product problems before choosing.
System depthBeyond default componentsRebuilt the visual language through branded tokens, typography, states, and interaction patterns.
Organizational resultTest space → production platformThe rapid experiment environment became the main place customer-facing product work shipped.
02 The full practice
One practice, three connected systems.
Prompty and evals are part of the picture. The broader work connects products, agent orchestration, model infrastructure, learning systems, and organizational change.
Experiencesets the need for Deliverycreates evidence for Learning
01Experience system
AI becomes a complete product experience.
I design the work before, during, and after the model, then give people and agents a shared design language that keeps the result coherent.
Product systemsDesign systems
Proof in practice
Dais + Cairn Adaptive feedback and structural intelligence
BENlabs system Branded components, standards, and API context
02Delivery system
People and agents share context, boundaries, and proof.
Ralph-style loops, specialist roles, planning records, quality gates, and explicit human decisions turn working proof into a repeatable operating model.
Agent systemsOperating systems
Proof in practice
Planning Harness + Prompty Shared plans, deployable loops, and guardrails
Builder portal AI-native experiments became production
03Learning system
Compare results against the same criteria.
I build tools that record what passed, what failed, and what changed between versions. Prompty Proof applies the same usability questions to both designs.
Evaluation toolsDesign review
Proof in practice
Prompty Proof Comparable runs and regression evidence
Witness · in development Annotation and training tooling; the first fine-tune on real annotations is planned.
03 / Across modalities
Many forms of AI.
One product discipline.
I build product experiences with text, image, voice, video, and 3D models.
Void Republic ComfyUI + image-to-3D workflows
From an image
to a 3D asset.
I built a ComfyUI workflow that generates five image candidates, grades each, and carries the strongest through 3D generation, mesh cleanup, texturing, and export.
Image generationEvaluation3D generation
Void RepublicOxbow freighter
Meshy · actual project asset01
Design reference
Reference
A reference image defines the form, proportions, and material language.
02
Geometry view
3D model
Inspect the form and the underlying mesh.
03
Texture
Materials
Reveal the color, panel detail, and wear on the same geometry.
04 / Prepared for the game3.05M 210kTriangles after mesh reduction
PreservedUVs + materialsThe visual detail carries through.
DeliveredUSDZIntegrated into Void Republic.
Oxbow · separate Meshy asset studyStudio views rendered from the project’s model.
Agentic product experiences
A workflow that
checks its own work.
Give it a brief. Watch it compose,
check, and improve a usable result.
Your brief
Prepare a launch composition.
Image + headline · finished editorial layout16:9 launch composition
Starting pointAutomatic fit
First attemptThe headline overlaps the subject.
Finished compositionA clear focal point. A readable headline.
Ready to useClear headline. Full artwork. Finished layout.
- 16:9 formatCorrect dimensions
- Whole formInside the frame
- Headline spaceClear of the image
The first layout misses the brief. The workflow adjusts the composition, checks it again, and leaves approval to you.
04 The connective tissue
Every project strengthens the next loop.
Give the work shape, context, and a bounded objective.
Choose the model, tools, roles, permissions, and handoffs.
Preserve outputs, behavior, provenance, and runtime evidence.
Compare the result against explicit checks and human judgment.
Improve the product, instructions, and next attempt.
Personalized feedback conversation → transcript + signals → tailored debrief → next attempt
Design review original + revised UI → same usability checks → comparison → next revision
Agent teams scoped goal → specialist work → quality gates → human decision → reusable context
05 A history of building
How my AI work developed.
- 2022–2023
Started building with generative AI
I began experimenting in 2022 and kept building personal projects through 2023.
- 2024
Tested what creator data could tell us
I coded small experiments to find out whether the available data was enough to produce useful insights for creators.
- Late 2024
Connected agents in AI-Sentinel
Before Claude Code, I built an orchestrator that passed each agent’s replies to the other as user messages so they could continue a conversation.
- 2025
Built a React Native proof in four weeks
I independently built a working mobile app with AI-assisted development. It helped the team reconsider what it could build with AI.
The four-week proof - 2025
Helped build the team’s product foundation
I led design and frontend work for a shared environment that became the team’s main place to build customer-facing features.
The shared product platform - Now
Building products through Labsly
Through Labsly, I’m building feedback products, agent workflows, and tools for evaluating AI output.
06 How I work
How I build and check the work.
Define the task.
I give agents a specific job and the context to do it. In Prompty, I keep reusable prompts and reference material together so I can bring that context into the next task.
Check the result.
In Prompty Proof, I compared a redesign with the original using the same five usability questions. I checked which problems were fixed and whether anything regressed.
Experiment with agents.
In late 2024, I built a prototype that passed one agent’s responses to another as user messages. I used it to explore how agents could work together.
Review the experience.
I design in working software and refine the interactions, accessibility, and edge cases. Automated checks help, but I still decide whether the product is ready.