Home on Andreas Nedelkos

You start out a new session and set the premise, ask some research questions, flesh out the architecture, and some hours later start building the first componenent.

Over time, ML models may degrade in quality as the data may gradually change, resulting in the model inference not being fit for purpose anymore.

When an edge device escalates a fault that exceeds its local reasoning capacity, the cloud tier takes over.

Standard tests check that components behave correctly under expected inputs.

Observing a deterministic service is straightforward: you instrument the hot paths, set thresholds on latency and error rate, and alert when something crosses a line.

Edge detection has to stay small because it runs close to the safety path.

Three edge agents and the execution runtime share state through CortexGuard’s Blackboard.

Anywhere software has to react faster than remote reasoning can arrive, you face the same problem.

How do agents operate in unknown, non-deterministic environments where the ground truth can shift from underneath their feet?