Client Work

Difficult products, built close to the people who use them.

I work directly with clients from an uncertain problem through design, implementation, and the less glamorous work of making software reliable. These examples are anonymous until each client approves being named.

Interactive construction planning

The problem: Residential plans contain a great deal of geometry and intent, but the source drawings were not made to behave like structured product data. People need to compare options, understand a design, and catch problems without becoming experts in the underlying file formats.

My contribution: I help build the path from complex drawing files to a rich browser workspace. That has included import and conversion, geometric data models, interactive plan views, option handling, publishing and recovery, and AI features that let people ask useful questions about a project.

The result: What began as difficult source material can become a repeatable, inspectable planning workflow. Plans can be brought into the product, explored as structured choices, corrected, and published again without throwing away earlier work.

Three-dimensional electrical design

The problem: Generated electrical layouts are only useful when designers can understand them, investigate problems, and make precise changes inside a complex building model.

My contribution: I work across the browser-based 3D application, from model loading and conversion through interaction and rendering. I have built and improved navigation, section cuts, transparent caps, outlines, cross-hatching, precise drawing and movement, constraint planes, snapping, orientation cues, and solution diagnostics.

The result: Generated conduit routes are not trapped in a static viewer. Designers can inspect them in context, edit them with construction-aware tools, and see clearer feedback about constraints and possible quality problems.

Semiconductor test software

The problem: A mature C++ and Python product for semiconductor engineers had to keep evolving without discarding years of code, build infrastructure, and domain knowledge.

My contribution: This was an advisory engagement, not implementation ownership. I reviewed architecture and live code with the engineering team, debugged parallel execution, built small examples to test design options, and advised on build and source-control tooling within the product's real constraints.

The result: The team kept ownership of production changes while gaining concrete designs, focused prototypes, and an outside technical perspective they could apply to scalability and maintenance work.

AI-assisted human resources

The problem: An AI assistant for human-resources work has to reach useful information and take useful actions without confusing companies, people, or provider records. A persuasive answer is not enough; the surrounding system must be trustworthy.

My contribution: I helped build document and employee workflows, realistic provider integrations, evaluation tools, and the tenant and identity safeguards that keep one organization's data and actions separate from another's.

The result: The accepted work gave the product a broader, testable HR data surface and reliable ways to exercise AI workflows against realistic provider behaviour, while keeping authorization and organizational boundaries explicit.

Field-service operations automation

The problem: A field-service company may depend on separate systems for customers, technicians, purchasing, accounting, messages, and documents. The same real-world entity can appear differently in each one, and a careless automation can create duplicates or write to the wrong record.

My contribution: I help build the platform and its guarded integrations. The work has included contextual company and people views, automatic imports with a human duplicate-review path, purchasing workflows, accounting synchronization, warranty research, and recovery for interrupted or ambiguous outside-system writes.

The result: Shipped workflows now bring operational records into one useful workspace and preserve where each fact came from. Repeats, retries, and possible duplicates are treated as normal product conditions instead of reasons to risk a silent bad write.

Additional Work

Custom mapping for a Squarespace site

As an unpaid favour for a friend, I built a custom Google Maps experience that helped visitors find participating restaurants and drop-off points, extending a Squarespace site beyond its standard tools.

If you are working on a product with this kind of technical depth, send me a note.