Resleeve, AI for Fashion · Ana Carolina de Carvalho Silva

Overview

Resleeve had real traction and a fast-moving team. The work was building the foundations to sustain it.

The platform AI tools for fashion creatives: text-to-design, sketch-to-photo, pattern blending, AI photoshoots, and brand style training.
When I joined Pre-PMF, high-velocity. Shipping frequently but without shared decision criteria or interaction standards to sustain the pace.
What was breaking AI models changed often. UI pivots were ad-hoc. Interfaces grew inconsistent, hard to extend, and engineering was accumulating debt each sprint.
My role Bring product direction and structure without slowing the team down. Deeply hands-on in feature development, using AI-assisted coding to build and ship POCs rapidly. Responsible for testing and evaluating new AI models to better achieve customer goals, and designing end-to-end experiences from concept through delivery.

Scope: Product strategy, AI workspace, onboarding, collaboration, approval flows, UI kit, AI model evaluation, POC development
Methods: Interface audit · Stakeholder alignment · User interviews · Usability testing · AI-assisted prototyping · Model benchmarking
Users: Independent designers, brand studios, enterprise fashion teams
Context: Pre-PMF · Fast-changing AI models · B2B scale ambition

The real problem

The team shipped regularly. But each sprint was partially undoing the previous one.

Symptoms No shared interaction standards. Interfaces inconsistent and hard to extend. Engineering revisiting the same decisions every sprint.
Root cause No clear criteria for what needed to stabilise vs. what could keep changing. Every decision was made from scratch.
The risk Resleeve was moving toward B2B scale on foundations too fragile to hold it. Each new feature cost more than it should.
Direction taken Separate the stable core from the experimental surface. Give the team scaffolding for fast iteration without the foundation breaking. Required alignment conversations with leadership first.

Work delivered
so far

AI workspace redesign Rebuilt the core interface around workflow state, reducing the navigation complexity that was blocking onboarding.
Credits system Defined a new credits system to allow quick changes to the credits deduction strategy, enabling different features to carry different costs as the business model evolved.
Payments & subscription flow Redesigned for B2B teams: shared credit pools, softer recovery when credits run out mid-session, and pricing tiers tied to actual operational costs.
Onboarding strategy In an environment of constantly changing models and priorities, aligned and created a customer support strategy to address LTV and retain customers.
UI kit Lightweight shared components to stop the team from making the same layout decisions from scratch each sprint.
Mobile adaptation Core workflow adapted for smaller screens, covering the main decision-making and review tasks.
v2.0 launch Beta with a group of enterprise customers, then full release. Coordinated across engineering, CS, and sales.
Discovery cadence Constantly observing user data, staying in direct contact with users, and gathering information for continuous improvements.

Outcomes.

2×

MRR growth within 6 months of v2.0 launch, driven by B2B account expansion and improved enterprise retention

40%

increase in weekly active usage within 2 months of v2.0 launch, driven by onboarding and workspace improvements

B2B

accounts onboarded onto the redesigned platform, validating product-market fit with enterprise fashion teams and brand studios

Learnings

→ Foundation work compounds. Once the team had clear interaction patterns, engineering stopped revisiting the same decisions and POCs stopped invalidating prior work. The delivery pace effect was noticeable but it required alignment conversations before it moved.
→ Name what needs to stabilise. Interaction principles and IA choices need more stability than surface features. Making that distinction explicit reduced rework and gave the team faster, clearer decision-making.
→ User trust can't be backfilled with features. In AI products, variable outputs and unpredictable latency are constants. Getting leadership and engineering aligned on treating error states and fallbacks as product requirements before they became support tickets was one of the clearest examples of where setting the right criteria early changed both the delivery outcome and customer retention.