AdvisorDX: Building an Industrial Diagnostics Product with an AI-Native SDLC
A joint case study by Cutsforth and Soliton Technologies
Abstract
AdvisorDX is a real-time diagnostic tool provided by Cutsforth for large industrial assets, such as generators, transformers, motors, and pumps. It combines several streams of sensor data (vibration, EMI, current signature, thermography, and partial discharge) into one verdict backed by evidence. The verdict tells a maintenance team which equipment needs attention, why, and how soon, in plain language.
Problem statement
- Expertise bottlenecks: Industrial sites already collect large volumes of condition-monitoring data from their critical equipment, but raw sensor data does not tell anyone what to do. Turning a vibration spectrum, an EMI signature or a partial-discharge trend into a diagnosis takes a vibration or monitoring-and-diagnostics specialist, and there are very few of them. Most customers have none on-site. So, maintenance teams either wait on scarce experts even for routine issues, or they miss early warning signs and learn about a failing asset only when it causes an unplanned outage. The data exists, but the expertise to act on it does not scale.
- Delivery bottleneck. Solving this meant building a scalable, multi-tier product: a web application, backend services, an analytics pipeline and cloud deployment, with enterprise-grade authentication, role-based access and multi-tenant data isolation. A small team could not build that fast enough through conventional development without letting timelines or quality slip.
Joint outcome
As a service provider, Soliton with Cutsforth, built an industrial condition-monitoring and diagnostics product through an AI-native software development lifecycle, with Claude embedded at every stage: requirements, design, feature breakdown, parallel implementation, testing and first-pass code review.
Solution
Introduced an agentic SDLC powered by Claude that covers requirements, design, build, testing and review. Domain experts set up the intent, and Claude carried most of the build under engineering direction and with human sign-off at every gate.
Stage | Claude product | What we did |
Requirements & planning | Claude Code, Claude (Slack) | Turned product intent into Jira stories, bugs and acceptance criteria written to a shared house style. Broke epics down into well-scoped stories and tasks. Summarized Slack threads and decisions into actionable work. |
Design | Claude Design | Carried product intent into screen designs, then straight into implemented, tested screens using the product's design tokens. The usual design-to-development handoff became one continuous flow. |
Build | Claude Code | Specialized sub-agents for frontend, backend services, database changes and analytics jobs worked in parallel across isolated work trees. Each one followed codified team conventions (skills and rules). |
Testing & QA automation | Claude Code | Wrote unit tests alongside every change and ran them until they passed. Authored and maintained the end-to-end Cucumber + Playwright suite against the running stack. Specialized agents audited accessibility, i18n and type safety. |
Code review | Claude Code (GitHub Actions) | Claude does the first review on every pull request and checks it against the ticket, the acceptance criteria and team conventions before any engineer looks at it. Security and correctness reviewers run on backend changes. |
Measurable outcomes
The baseline covers the eight months before the team adopted Claude across the lifecycle (Oct 2025 – May 2026). The “With Claude” column covers the four months since (Jun – Sep 2026).
Metric | Baseline | With Claude | Change |
Merged pull requests per month | 50 | 227 | +354% |
Commits per month | 54 | 237 | +339% |
Automated test files (unit, integration, E2E) | 158 | 727 | +360% |
Commits co-authored by Claude | ~1% | 62% | +61 pts |
PRs with a first-pass AI review | 0% | 100% | Every PR |
Story cycle time, weighted by story points (days)* | 7 | 3.5 | −50% |
Mockup ideation | One design direction per iteration | Several directions explored in one session | Faster ideation |
* “With Claude” means most of the story’s PRs were co-authored by Claude; the baseline is the remaining stories from the same period.
Conclusion
- Engineers direct intent instead of writing code. Agents drive the whole path from requirement to a production-ready feature, and engineers set direction, review and approve.
- Review effort dropped. Claude catches convention, test-coverage and correctness issues first, so human reviewers spend their time on design and domain questions.
- Quality held while speed went up. Every change ship with tests, a first-pass review and the full lint, test and build gate.
This case study was co-authored with Cutsforth Inc.
Cutsforth is an industrial condition monitoring and reliability company with more than 30 years of experience in generator systems worldwide. Cutsforth has served more than 1,000 customers in over 20 countries.
Industrial Condition Monitoring Solutions | Cutsforth