Context before analysis
A message does not exist in isolation. Prior exchanges and relationship dynamics change what a useful response looks like.
A calmer way through
difficult messages.

The premise
When a message carries conflict, urgency and history, the hardest part is rarely typing a reply. It is seeing the situation clearly enough to choose one.
Metis is an AI communication assistant for people navigating difficult separation and co-parenting conversations. It helps them pause, inspect an incoming message and draft a response with less escalation risk.
The product is deliberately narrow. It does not diagnose people or replace legal advice. It creates a small layer of composure between a provocative message and the answer that follows.
01 / Product principles
A message does not exist in isolation. Prior exchanges and relationship dynamics change what a useful response looks like.
The product points to observable words and patterns. It does not pretend to know what another person is thinking.
De-escalation should preserve boundaries and agency—not ask the user to accept bad behaviour.
Sensitive context stays in the browser wherever possible, with clear limits around what the product can and cannot do.
02 / The product

Metis separates observable patterns, emotional pressure and practical risk, then offers a grounded next step.

Instead of manufacturing one “perfect” reply, the product makes the trade-offs legible: reduce tension, hold a boundary or keep the record clean.

A private local history helps patterns emerge over time while keeping sensitive communication close to the user.
03 / Early signal
An early distribution test—not a victory lap.
What the work proved
High-stakes AI does not need more confidence. It needs clearer boundaries, better context and the discipline to leave judgment with the person.Back to selected work