Adopting a philosophy of simplicity by using a single source of truth and composable tools reduces operational complexity for small teams.
Using single-purpose tools like Nomad and Consul provides most of the value of a larger platform like Kubernetes with a fraction of the operational overhead.
Leveraging Postgres with PostGIS and Neon provides a unified data layer that avoids the complexity of synchronizing multiple specialized databases.
A Kalman filter for cleaning noisy GPS data can be implemented as a PL/pgSQL function, eliminating the need for a separate stateful stream processor.
Using database triggers and continuous queries to merge location points into trips provides near-real-time results without an external stream processor.
A temporal graph of vehicle-trailer relationships can be modeled and enforced directly in Postgres using interval columns and deferred triggers, avoiding a separate graph database.
Offloading old data to Parquet on S3 and querying it from the client's browser with DuckDB-Wasm eliminates the need for a server-side data lake.
The simplified architecture has trade-offs like a steeper learning curve for PL/pgSQL, but the benefits of reduced operational complexity are significant for small teams.
Jobs that call for the skills explored in this talk.