I find the problem, then I ship the product.

How we built it, in two minutes

From the first assumption to the timeline, step by step: the interviews, the synthesis, the personas, the journey, the vote, the user flow, the prototype, where the AI does the work, and the architecture. Pause it, drag the timeline, or jump to a step.

The problem

Furnishing a new place takes months: prices checked tab by tab across IKEA, Wayfair, Amazon and Facebook Marketplace, guesses about whether a piece will fit, and no designer to ask. Existing tools either stay inside one retailer's catalog, rely on slow human designers, or generate a pretty room with nothing you can actually buy.

What the research said

Our team heard from 30 people furnishing a home, in 18 live or voice-note interviews and 12 form responses, and coded every answer into 36 pain-point themes.

Homie voice-of-customer research, 30 people, Oct 2026 · cart abandonment: Dynamic Yield via Oberlo, Oct 2024

The pitch deck

From 30 interviews to a clickable prototype, in 29 slides. Click the right or left side of a slide, use the arrow keys, or swipe.

01 / 29 Homie: your personal interior design expert

From pain points to answers

Who we build for first

We sorted the 30 people into the team's six persona types and design for three of them. We start with Budget New Movers like Maya, 13 of the 30 and the largest group: she starts from an empty room, can't paint or drill, and price leads every decision. Existing Furniture Optimizers come next and High-Investment Planners later, as a premium tier.

What I built

• Homie appAn AI interior designer for people moving into a new home, built as a clickable web prototype: 22 screens across the whole journey. You tell it about the room and send a photo; it restyles your own room with real, buyable pieces, gives one pick per piece with the reason and the price across stores, checks each piece against your measurements, and ends in one cart. Every screen traces back to what the 30 people asked for.
• AI hypothesisIf a mover sees her own room restyled with real, buyable pieces and a reason for each pick, she can approve and buy in one session. We earn retailer commission on each order.
• PersonasThree personas sized from the 30 interviews, and the case for starting with Maya, a budget new mover.
• Pain-point mapAll 36 pain points the 30 people raised, placed on the six stages of furnishing a home, each traced back to the answers it came from.
• Lifecycle planA concierge proof of concept first, then an automated one, a pilot with real movers, and an MVP gated on movers actually buying.
• Room Try-OnThe building block behind Homie's fit check, built as a Shopify app for decor stores. Shoppers upload a room photo and drag, resize and rotate the product onto their wall. Product photos are cut out in the browser, so nothing leaves the device. An optional “Render with AI” button sends the mock-up to Gemini through a signed app proxy with a monthly cost cap.

Also produced

The problem

A large share of a product manager's week goes to writing the same kinds of documents: vision statements, PRDs, personas, journey maps, interview guides. Each starts from a blank page even though the structure is well known.

What I built

A tool library that drafts each artifact from the product's own context, using retrieval-augmented generation so outputs stay grounded in what the team has already written. Drafts open in a built-in editor for the PM to refine.

• Vision & strategyProduct vision statements and PRDs
• DiscoveryPersonas, customer journey maps, customer interview guides
• ExperimentsLean startup hypotheses and experiment design
• PrioritizationRICE scoring and use-case generation
• Go to marketProduct marketing copy and LinkedIn posts

Business model

Three monthly plans from $10 to $50, each with a monthly word allowance and access to every tool, plus a free trial with no card required.

The problem

People don’t have a recipe problem; they have a saved-but-never-cooked problem. Hundreds of recipes sit in a saved folder while the same few dinners repeat, because turning a video or a page into a plan and a shopping list is work nobody has time for on a weeknight.

What I built

• Save from anywherePaste a link to a cooking video, a recipe site or a blog, or photograph a cookbook page or a handwritten card. AI reads it and captures the ingredients, steps, timing and nutrition.
• A week that plans itselfIt builds the week from your own saved recipes, works around busy nights, and says why each meal is there. One tap swaps a meal.
• The list writes itselfEvery ingredient from every planned meal, merged and sorted by aisle, shared with the household.
• It learnsRatings shape next week’s plan; allergies are hard rules, never suggested.
• No loggingOptional calorie and protein targets, tallied from the plan. Step-by-step cook mode with big type.

Getting through App Store review

The first submission came back with three issues: accounts could be created but not deleted, the screenshots showed a drawn status bar and other platforms’ names, and the reviewer asked where subscriptions are bought. One new build fixed all three: in-app account deletion that removes the user’s data first and the account last, so a half-failure never strands anyone; clean screenshots and generic source names; and a plain answer that the app is free, with no purchases at all.

The problem

Patients struggle to get their records from different providers, to remember their own medical history, and to share it, including family and dental records. In Canada, 44% of people see more than three providers.

What the research said

A survey of 29 people across provinces, plus 5 interviews with patients and providers.

Survey n=29 · 5 interviews

The solution

Patients set the wallet up around their own health needs, see their history by provider, and choose exactly what to share. Each provider gets a QR code or access code to open the shared record. Integration with provincial and private health systems runs through a privacy-compliant API, and patients pay a small monthly fee.

Also produced