Engineering Excellence with Claude

At a Glance

PixieDust AI is building the AI-native operating system for small and medium business (SMB) lenders and merchant service providers in the US lending market. As PixieDust's engineering partner, Ryewalk has been part of this journey in building an agentic platform - transforming an Idea to a fully working product, with the following features:

The engagement's focus was to deliver a full-working AI-led, agentic solution for PixieDust to accelerate the process of lending application qualification and increasing the conversion rates for credible loan portfolios.

This Claude based AI-engineering engagement runs as an agentic software development process applying Claude Code and Claude skills across architecture, implementation, code review, documentation and operational runbooks by Ryewalk's team of Claude-certified architects. In other words, the same AI technology that powers the product is also used to build and scale it.

The impact for PixieDust is immediate and tangible:

Client Background & Business Vision

PixieDust is a US-based AI-native operating system for SMB lenders and merchant service providers, founded by experts from PayPal, Funding Circle, JPMorgan, FICO and Dun & Bradstreet — the people who ran the lending infrastructure are now rebuilding the same with an AI-first approach. Their customers are lenders and merchant service providers serving small businesses with working capital, revenue-based finance and Small Business Administrator (SBA, US regulator for SMB lending) aligned products. PixieDust is trying to help them turn more qualified SMB leads into funded loans, without proportionally increasing their sales and operations headcount.

While PixieDust leadership brings unmatched industry and domain expertise, they needed a strong engineering partner in Ryewalk to convert their vision and idea into a workable solution.

The Engagement

PixieDust's target problem is fragmentation of the application process rather than the actual credit roll-out itself: roughly half of qualified SMB borrowers never finish an application because follow-ups lapse, failed banking connections and applications stall part-way through. No CRM or loan origination system owns that stretch of the funnel.

But PixieDust's agents do. These agents engage the borrower, run the qualifying conversation, collect what underwriting needs, and hand the lender a complete application rather than a lead.

Two agents run in production:

While PixieDust owns the product, the lending domain and the customer relationships, Ryewalk owns the engineering execution end-to-end. The platform is live across five lender brands, each with its own products, credit box, agent personality and approved material, running from one configurable base rather than five separate builds.

The Delivery Challenge

An early-stage company launching into regulated lending has two clocks running against it:

That combination usually forces a trade. Teams either move fast and carry the compliance debt or build carefully and arrive late. PixieDust could afford neither. The platform had to launch, then scale to additional lender brands without the engineering cost multiplying by brand, all while a small team held the roadmap. This needed a scalable technology architecture with execution capabilities, decoupling business expansion with cost of technology operations.

The pressure landed on the engineering side of the engagement. Ryewalk had to deliver a regulated-grade platform at startup pace, then keep the marginal cost of each new brand close to zero.

How Ryewalk Delivered

Ryewalk ran the engagement as an agentic SDLC rather than a conventional one. Claude Code and Claude skills were used across the lifecycle by Claude-certified architects, with a person deciding every merge.

1. 01
   Claude across the lifecycle.

   Architecture and technical design, implementation, code review, documentation and operational runbooks all ran through the same tooling. The intent was that Claude carries the execution while engineers move up to specifying and reviewing, which is where the leverage sits on a small team.
2. 02
   Routine work stopped being expensive.

   Boilerplate, test coverage, refactoring and migrations are the work that consumes a delivery team's week without distinguishing the product. Moving that class of work through Claude Code is where the cycle time reduction came from and it compounds: the time it returns goes back into the work only the team can do.
3. 03
   Standards were versioned, not remembered.

   Rules, skills and review standards were treated as shared assets installed consistently, so that the same standard applied to every session, rather than depending on who ran it. New engineers joined into an established practice instead of rebuilding habits from scratch.
4. 04
   Configuration replaced rebuilding.

   The architecture keeps Claude reasoning while deterministic code owns every action touching a lender system, with a full audit record. Lender behaviour is a configuration rather than code, so that the material that the compliance team approves is the same material the agent is held to, and adding a brand is only a mere configuration exercise rather than a full-out build. That is the reason why decisions for subsequent brands did not cost the same as the first brand.

What It Meant for PixieDust

For an early-stage company, engineering velocity converts directly into runway and market position. Reaching production sooner meant the platform was in front of lenders sooner, and the multi-brand rollout followed without a proportional increase in engineering spend.

The capacity returned by faster routine work went back into the product. Time not spent on migrations and test scaffolding was time spent on the lending logic, the conversation design and the compliance surface, which is the work PixieDust's domain expertise actually bears on.

As the engagement continues, Ryewalk operates and extends the platform under an ongoing capacity model, onboarding new brand configurations and deepening production usage as PixieDust's roadmap moves.

Customer Perspective

Measured Outcomes

Both figures are Ryewalk's own delivery measurements, taken across the launch and multi-brand scaling phase and benchmarked against comparable engagements.

The last two are the most durable. Time to market is a one-off win on a launch. A halved cycle time on routine work, and a brand rollout that costs a fraction of the first one, are permanent changes in what a team of a given size can carry, and they persist into the operate-and-extend phase.