Overview
The project began with a central tension in Cognitive Behavioral Therapy: many emotional patterns only become visible over time, but therapy often relies on fragmented memory, irregular sessions and subjective descriptions.
Rather than treating AI as a replacement for therapy, the project investigates how an interface could help users notice patterns, reflect on cognitive distortions and access small moments of support in everyday life.
My role focused on translating the concept into an interaction system and working prototype. I worked on the conversational flow, AI behaviour, technical implementation, emotional feedback logic and the connection between voice, chat and dynamic visual states.
A local-first AI system that combines conversation, emotional analysis and biometric context.
Context
Mental health support is often limited by cost, availability and continuity. Even when people receive support, much of their emotional life happens outside the therapy session — during work, conflict, isolation, stress or routine moments that are easy to forget later.
Sky was developed as an AI-supported CBT companion that works closer to the user’s real environment. It combines voice interaction, affective computing and personal data widgets to create a more continuous form of reflection.
The project asks how sensitive technology can become more adaptive without becoming invasive. This made privacy, local processing and emotional transparency central parts of the concept.
Research
We focused on a central gap in digital mental health tools: many systems ask users to reflect, track and report their emotional state, but often do so in moments where users are already overwhelmed, distracted or unwilling to engage with another interface.
It began by examining Cognitive Behavioral Therapy as a framework for self-awareness, reflection and self-regulation.
This helped define Sky not as a diagnostic tool, but as a system that supports users in recognizing patterns, questioning automatic thoughts and returning to their own agency.
UML Over User Flows:
Sky could not be mapped through traditional user flows alone. The AI acts as a co-agent in the system, responding to emotional context, risk signals and changing user needs.
We used UML diagrams to map these decision paths more clearly, including therapist interventions, emergency escalation and API handoffs. This helped define where the system responds, where it pauses, and where responsibility shifts from AI to human support.
How can technology support users without becoming invasive?
How should the AI system
communicate?
Through voice mapping, feedback sessions and interaction testing, we defined a conversational behaviour that avoids overload. Sky should ask one question at a time, respond with empathy before offering direction, and adapt its tone without becoming judgmental or overly familiar.
The research also questioned whether a mental health interface should always be screen-based. Several explorations focused on diminished GUI, audio-first interaction and physical entry points.
This led to the idea that Sky should be available in moments of natural openness — during a walk, at night, after work or in quiet self-care routines — without forcing users into a dashboard.
Sky’s orbs turn emotional states into a visual language of colour, rhythm and motion — from calm listening to deeper analysis.
Feedback throughout the project shifted the concept away from a general wellness app for everyone.
Instead, Sky became a more specific exploration of how AI could support psychological awareness without creating dependency.
The goal was not constant optimization, but a system that helps users understand themselves better and eventually need the system less.
Sky is with through voice, reflection and adaptive feedback.
Technical SystemSky was built as a working MVP with a FastAPI backend and a web-based frontend using Vanilla JavaScript and React components.
The prototype connected conversational input with CBT-based response logic, emotion analysis and dynamic visual feedback. User messages were processed for emotional tone, cognitive patterns, conversation topics and crisis signals.
These outputs informed both the therapeutic response and the behaviour of the visual interface.
Some features were implemented directly in the prototype, while others remained part of the technical vision, including full Whisper speech-to-text, ElevenLabs text-to-speech, user accounts and deeper biometric integration.
The system also explored speech-oriented interaction, facial emotion recognition, ambient noise detection and multilingual support.
The technical goal was not to build a finished medical product. It was to create a believable prototype of an AI-supported reflection system where conversation stays in the foreground and analysis happens quietly in the background.
Sky is built around a three-screen interaction model: conversation, body and mind.
Interaction SystemThe main screen is the conversational space. It is where users speak or type naturally and where Sky responds through a calm, step-by-step dialogue.
From the main screen, users can swipe horizontally into two supporting layers.
One side opens the body layer, where biometric and contextual signals become visible. This screen connects emotional reflection with information such as sleep, heart rate, movement or routine changes. It helps users understand how their mental state may relate to their physical and everyday context.
Swiping in the opposite direction opens the mind layer. This screen contains insight blobs, recurring themes and emotional patterns that Sky has identified over time. These insights are not presented as diagnoses. They act as reflection points that users can return to, question and understand at their own pace.
The swipe structure keeps the system spatially simple. Conversation stays in the centre. Body-related context sits on one side, and emotional reflection sits on the other. Users do not need to search through menus or manage complex navigation. They speak first, then move sideways when they want to understand more.
Across the system, the Sky blob acts as the central feedback element. It reacts to voice, rhythm and emotional intensity through movement, colour and scale. The blob gives the system a sense of presence without turning it into a human-like therapist avatar.