← Selected work

AI product / Human-guided agentic SDLC

NearHalo: A human need. An agentic product team.

I turned a personal need into a product and used specialist AI agents across discovery, design, engineering, testing, and launch. The direction, decisions, design quality, and accountability remained mine throughout.

The film / 3 min 18 sec

The product story and the team behind it.

Press play to watch the full landscape film. Closed captions are available in the player; the complete narration is below.

Full film · landscape 16:9 · published October 7, 2026 · audio on demand · Open the video file

01 / The human problem

Independence and a clearer way to know when help is needed.

My father suffered a stroke while living independently. What stayed with me was how long he had been alone without any of us knowing he needed help. I wanted to support his independence while giving family a better way to recognize when a check-in might be needed.

NearHalo began with that need, not with a technology demo. The film tells the personal story in my own words.

02 / My role

Product judgment stays with the person accountable.

I set product direction, supplied the problem context, worked through options with product agents, guided design consistency, and made the decisions that carried into implementation. The agents extended my capacity; they did not replace the judgment behind the product.

For new features and enhancements, agents debate options and refine a recommendation. I decide whether to proceed; when I approve, a lead agent implements it through the connected harness. That is one explicit decision point, not a claim that every action at every stage passed a formal approval gate. I remain accountable as the work moves from research to launch.

03 / The workflow

Context, specialists, review, and iteration.

  1. Frame the need. BMAD was the foundation. I added specialist agents and reusable skills. Through context engineering, product agents questioned the pain point, proposed options, and worked through choices with me; the decisions became shared context.
  2. Research and shape. Research agents compared products and identified gaps, shaping personas, journeys, and flows. My design loop kept interactions and visual language consistent and flagged generic AI styling.
  3. Build and challenge. Engineering agents implemented changes. Independent review agents flagged code and potential security issues; engineering agents made fixes and regression tests checked them.
  4. Check the experience. Alongside the native app I built a web app. Agents used Playwright to check browser flows for usability friction and ran automated accessibility checks.
  5. Carry context to launch. The established product context informed the marketing website, including captured product screenshots and UX copy grounded in personas. The story stayed connected from problem framing to public explanation.

Workflow in practice / 1 min 25 sec

Human review needs a user experience.

To make human review usable within this workflow, I designed an interim test-review interface for NearHalo. It organizes 37 checks across the Caregiver app, Recipient app, and Apple Watch.

Device-specific steps and expected results clarify what to check. Pass, Problem, or N/A choices and notes make my assessment explicit, while Jira references connect each case to the underlying work.

This walkthrough demonstrates the review design, not completed test results: the recording shows 0/37 checks complete. The checkpoint complements fuller testing; human judgment remains part of the workflow throughout.

Read the walkthrough transcript

Human review needs a user experience, not just a checkbox.

NearHalo is a complicated product, so I added a layer of governance: BMAD agents automatically create stories in Jira through Jira MCP.

That gives me a structured place to review the stories, tasks, and their relationships.

I also use the timeline to see how the work is organized across sprints.

Testing spans the Caregiver app, Recipient app, and the recipient’s Apple Watch app.

Some checks can be overlooked. So I designed an interim checkpoint, with thirty-seven checks to clarify what to test.

Each Claude Code question names the device, its test steps, and expected results.

I can mark Pass, Problem, or Not Applicable, and leave notes about what I actually saw.

Filters help me navigate the list, and Jira references keep each case connected to the work.

This complements fuller testing. Human judgment still matters throughout. For me, that's a product-design problem. How are you designing the experience of reviewing AI-assisted work?

Human-review walkthrough · 1:25 · audio on demand · English captions

04 / Outcome and evidence

A product beyond the prototype.

I took NearHalo from initial ideation to App Store readiness in an estimated 15 full working days of development effort. Specialist AI agents supported the lifecycle, while I retained responsibility for product direction, design decisions, review, and approval. The app subsequently became available on the Apple App Store.

The film presents a native app on the Apple App Store, a web app, and a marketing website. Separately, the public App Store listing, product website, and web-app entry point are currently reachable; their availability does not verify every feature or user outcome shown in the film.

Automated browser and accessibility checks are not comprehensive human usability or accessibility evaluation. Independent code and security review can surface issues but does not guarantee security. NearHalo supports family care coordination and general wellness; it does not diagnose or prevent stroke, guarantee rescue, or replace medical care or emergency services. Information and alerts can be delayed, incomplete, or unavailable.

Private Access

Want the deeper work?

Discuss my design leadership, methods, and publicly shareable work in a focused hiring conversation. Confidential client information is not shared; additional materials require explicit authorization for disclosure.