Product Lab / nearbyto-me
NearbyTo.me: building safety into an AI product
A Product Lab record about a self-owned experiment, its safety controls, and the evidence boundary around an unresolved trial status.
Product relationship
This is Oleg's own product and experiment, not client consulting proof.
Maturity
The trial status is unresolved. Public efficacy conclusions are excluded.
Claim boundary
The trial status remains unresolved. This draft does not claim efficacy, completed enrollment, or client validation.
Safety constraints
The source package records the public surface, full-product scope, evaluation framework, and prioritized safety findings as separate facts.
- The public web surface took six weeks; the full product took approximately 12 weeks.
- The evaluation and observability framework contains 30+ metrics.
- P0/P1/P2 identify safety findings and remediation priority, not efficacy.
Result type
A self-owned product-building observation. It is not a client result or a public production metric.
Experiment record
A consequential AI product needs inspectable safety controls before its results can carry commercial meaning.
I worked on NearbyTo.me as a founding AI product architect. This record covers product-building evidence, not a consulting engagement.
The product had to keep a useful conversational surface while making risk detection, interruption, and observability explicit.
Separate product maturity, safety evidence, and trial evidence before using any of them as proof.