I help shape digital products

Between a surgeon's order and the operating room there is a two-to-three-month process, run by one coordinator over WhatsApp and a spreadsheet. A third of 2,500 historical cases had no recorded outcome. The funnel was not leaking. It was invisible.

I rebuilt the baseline from the source data before designing anything, then audited my own numbers and cut the business case by 40%. I mapped the journey with the coordinator and turned it into a state machine. I defined what the system never does alone: talk to doctors, read handwritten orders, communicate a suspension. Autonomy is earned with a metric, 70% of drafts approved unedited, not declared.

First interview to production in four weeks. Runs alongside a live service with four surgeons and 150 to 200 open cases. Measured 45-day pilot with a control group from October 2026.

A lead-to-project matching engine for real estate developers had to serve two businesses at once: a white-label product for each developer, and a broker channel where we capture leads and lend them out. The proposed data model treated ownership, management and origin as one thing.

I tested the product myself and isolated the real failures. I rejected the model in writing, with three contradictions and six edge cases, and redefined the domain around three relationships. That cut the change from over 3,000 lines to under 900. I ran the decision meeting in 34 minutes with three yes-or-no questions, then specified the new screening flow and the analytics to judge it.

Multichannel model, attribution and metrics in production one month after the spec. Catalog of 38 developers and 138 projects. Now in validation.

A B2B language-learning platform with around 800 active students, 100 groups and 25 teachers was running enrollment, grouping and teacher payroll by hand. Its community features went unused for reasons nobody had named.

I diagnosed the community problem as psychological, not technical, and reframed it from blog to plaza. For the backoffice I sat with the people doing the manual work and scoped by pain: reduce 80% of it, not 100%. The system suggests, a person approves. Ten specs with a closed vocabulary, acceptance criteria and an explicit "not included" list. 275 decisions tracked in four weeks, 21 reversals, 18 of them my own proposals rejected.

Operations backoffice specified and prioritized in four weeks. First sprint scoped to what unblocks moving 500 people onto the new system.

Canoe and rowing clubs in Rio de Janeiro fill their boats through a WhatsApp list copied by hand. I parsed 9,000 messages from two clubs: 83% were copies of the list, and 30% of cycles lost confirmations to concurrency. The clubs had no idea. The problem is invisible from inside.

I built the evidence, corrected my own occupancy number when it did not match the club's prices, and sized the market with a manual census. I designed a link that lives inside the group instead of an app. Rules: the coach approves, never drafts; nothing about the sea gets automated. I priced it, planned a gated pilot, and wrote two different decks for the club and the partner.

Pre-pilot. Brand, prototypes, and a plan that can stop without debt. Also the reason I row at 5:40 in the morning.