Studova

The core problem: research studies have wildly different shapes. A cognitive test, a survey battery, and a longitudinal tracking protocol share almost no structure. The platform had to enforce reproducibility without constraining what a researcher could design.

Three decisions defined the architecture.

Graph-based flow engine. Study workflows aren't linear; they branch, randomise, and loop. I implemented a directed graph using FlowElements and FlowLinks, with a traversal algorithm that validates path integrity from START to END before a study can launch. Any broken flow is caught at design time, not mid-session.

Dynamic schema validation. Each task type collects different data. A rigid relational model would require a migration for every new task. Instead, I combined typed ResultVariable models with PostgreSQL JSON columns and GIN indexing: type safety and query performance without the migration overhead. Zod schemas are generated at runtime from the stored metadata.

Results visualisation with accessibility. Displaying participant progress across arbitrary task sequences required ordering columns by flow position, not insertion order. I built a colour-coded and pattern system (colour for completed, hatched for pending) so the view works for colour-blind researchers. Small detail, high stakes in a clinical context.

It proved that a solo engineer can ship a production-grade research tool, if the architecture decisions are right from day one.

An early-stage SaaS I'm building for research-study management.