A workshop,
not a portfolio.
EmergeOS
A cognitive performance platform that reads across your own writing and shows you the loops you can't see from inside.
emergeos.appMost people who journal never see the patterns in their own thinking, because seeing patterns requires reading across dozens of entries over months and holding all of it in mind at once. That is not something a person can do for themselves from the inside. EmergeOS reads across the writing, surfaces the loops and distortions that keep returning, and turns each one into a structured behavioral experiment with a measurable outcome.
Cognitive distortions — Beck
14 canonical distortion types — catastrophizing, mind reading, all-or-nothing — detected with confidence ratings and direct evidence quoted verbatim from the user's own writing. Not summaries.
Memory reconsolidation
Beliefs become plastic when reactivated. Experiments record a pre-belief rating and a prediction, then a post-outcome log — the platform tracks whether the underlying belief shifts across weeks.
Predictive processing
Loops are explained mechanistically — what the brain is predicting, what it gets wrong, how the error reinforces the next prediction — rather than reduced to a generic label.
Self-Determination Theory
Autonomy, competence, and relatedness are scored from journal content with grounded reasoning and tracked over time. Unmet needs predict motivational collapse — the system surfaces those early.
Circumplex model of affect
Mood is captured in two dimensions — valence and arousal — instead of a five-point scale. Activated-negative (anxious) and deactivated-negative (depleted) are different states and treated as such.
A three-stage analysis that cannot output anything ungrounded in the user's own writing.
Condense
RAG selects the most relevant entries by embedding similarity and reduces them to a structured brief: themes, emotional arcs, candidate evidence.
Analyze
Named distortions with confidence and citations, self-reinforcing loops with each stage articulated, unstated tensions, SDT scores, and targeted behavioral experiments.
Validate
A separate AI pass checks that every cited quote appears verbatim in the source text, that patterns are clinically accurate, and prunes anything generic.
On-device ONNX crisis detection across English and Estonian. Safety rules hardcoded into every system prompt — not configurable. Row-level isolation at the database layer regardless of application logic.
I'm 19, based in Estonia.
I build tools at the intersection of cognition and automation. Most of what I make, I make for myself first.
That's the only way to know if it's real.