Selected projects

The situation

Pulseboard’s original stack had been assembled during an early fundraising sprint. It worked — until enterprise customers asked for stricter tenancy isolation, longer retention windows, and dashboards that stayed responsive during peak ingest. Query latency climbed, nightly jobs overlapped with daytime traffic, and the product team spent more time explaining outages than shipping charts.

What we did

We started with a joint discovery week: mapping the highest-value dashboards, the tenants that generated the most load, and the contracts that could not slip. Then we split the work into three tracks — product UX for the densest screens, an API boundary that could survive a storage migration, and a phased move of analytical workloads onto a warehouse path without a hard cutover.

Frontend stayed on Next.js with a tighter design system so operators could scan anomalies faster. Backend services were reshaped around clearer tenancy middleware, rate limits, and observability that product and engineering could share. The customer-facing IA barely moved — which was intentional.

Outcome

Pulseboard reached 180k DAU on the new path with p95 dashboard loads under the targets agreed with their largest accounts. The internal team took full ownership after an eight-week overlap, with runbooks and a staging environment that mirrored production tenancy.

DAU growth

2k → 180k

p95 dashboard

< 1.8s under peak

Handoff

8-week overlap

The situation

Harbor Ops connected shippers with regional carriers. The marketplace existed as a collection of spreadsheets, a fragile quoting tool, and a scheduling board that broke whenever peak season arrived. Sales could close deals; operations could not fulfill them with confidence. Leadership needed a single system that matched how dispatchers actually worked — not a generic CRM with logistics labels.

What we did

We embedded with ops leads for two weeks of ride-alongs and desk shadowing. That research shaped a quoting flow that surfaces capacity constraints early, a scheduling surface optimized for keyboard-heavy work, and an operator console with clear ownership of exceptions.

Engineering focused on idempotent booking APIs, Redis-backed locks for contested slots, and background workers that could absorb burst traffic without dropping carrier updates. React handled the dense operational UI; Go services owned the transactional core.

Outcome

Harbor Ops launched region by region. Quote-to-book time dropped, exception queues became measurable, and the marketplace survived its first peak season without rolling back to spreadsheets. The internal team now extends the platform with our design tokens and service templates.

Quote → book

−42% median time

Peak season

2.1× volume, stable SLOs

Regions live

6 in year one

The situation

Fold Finance needed to move beyond a founder-built prototype before onboarding regulated partners. The customer app and internal admin had drifted apart — different patterns, incomplete audit trails, and auth that would not pass a serious review. Speed still mattered: a partner pilot had a fixed start date.

What we did

We unified both surfaces under one design system and TypeScript monorepo. GraphQL became the contract between product UI and domain services. Auth moved to a managed identity layer with MFA, session policies, and role-based access that ops could reason about. Every sensitive action wrote an immutable audit event.

UX work focused on reducing cognitive load in money movement: clearer statuses, reversible steps where safe, and admin tools that made support investigations minutes instead of hours.

Outcome

Fold cleared its partner security questionnaire on the first submission cycle, launched the pilot on schedule, and kept a single product language across customer and admin experiences. Their team continues development on the same foundations.

Security review

Passed first cycle

Support time

−55% on payment inquiries

Pilot

On-time launch

The situation

Northline’s operations lived across email, a ticketing tool, and a custom spreadsheet that had become a de-facto database. New hires needed weeks to learn the ritual. Leadership wanted visibility without another “status meeting.”

What we did

We mapped the real workflow — including the unofficial steps people had invented to stay sane — then designed a single queue with roles, SLAs, and automated handoffs. Temporal orchestrated long-running processes; the UI stayed deliberately quiet so operators could scan queues all day without fatigue.

Outcome

Time-to-productivity for new ops hires dropped from weeks to days. Leadership got reliable throughput metrics for the first time, and the spreadsheet was archived with a small farewell thread.