Experimentally Rigorous, Built to Validate
Novel perception systems require novel validation. It's not enough to build something that works in a demo — you need evidence that it works for real users, in real conditions, compared against meaningful alternatives. We design systems with experimental comparison built in from the start.
Every system we build is instrumented for data collection, structured for controlled comparison, and ready for the kind of rigorous user studies that produce publishable results and defensible design decisions.
What We Build
Multi-Strategy A/B Testing Frameworks
When you're exploring a design space — different sonification approaches, different depth representations, different interaction models — you need to compare them under identical conditions. We build testing frameworks that let you switch between strategies in real time, with matched stimuli, counterbalanced ordering, and automated condition management. Same hardware, same environment, same participant — different approach.
Real-Time Telemetry Systems
Every interaction generates data. We instrument perception systems to capture performance metrics (latency, frame rate, model confidence), user behavior (navigation patterns, response times, error rates), and environmental context (lighting, noise level, movement speed). This telemetry streams to structured storage in real time, enabling both live monitoring and post-hoc analysis.
User Study Infrastructure
Running user studies with assistive technology requires careful infrastructure — especially when participants include people with visual impairments. We build study platforms with accessible interfaces, screen reader compatibility, audio-first interaction patterns, and configurable consent and debriefing flows. The technology handles the logistics so researchers can focus on the science.
Publication-Quality Data Pipelines
The path from raw telemetry to a figure in a paper is long and error-prone. We build data pipelines that automate the journey — cleaning, aggregation, statistical testing, and visualization — with reproducible scripts that generate publication-ready outputs. Every analysis is version-controlled, every figure is regenerable from source data, and every statistical test is documented.
Research Philosophy
We believe that building and evaluating are not separate phases — they're the same process. A system designed for evaluation produces better engineering decisions throughout development. When every design choice will eventually be tested against alternatives, you naturally build more modular, more configurable, more measurable systems. The research makes the engineering better, and the engineering makes the research possible.