Home of the Uedu platformShow a desktop notification when the AI TA finishes replying
Notify me when classmates post messages in the forum
Play an alert sound whenever there is a new notification
Physiologically-Aware Language Models — let the AI TA understand your physical and mental state. When you start a conversation, the system will automatically adjust the AI’s response style according to your sleep, stress, heart rate variability and other physiological data.
Illustrative · dummy data
Every advance in educational technology expands AI's understanding of learners
Digitising content
AI knows the Student's answers: whether they were right or wrong, and how long they took. Learning extends from the classroom to the screen.
Devices and mobility
AI adds device awareness: screen size, network environment, interaction mode. Learning is no longer tied to desktop computers.
Environmental context awareness
AI knows the environment: location, time, temperature, social context. The learning situation begins to be understood.
Embodied context awareness
AI finally senses the learner's body: autonomic nervous system state, sleep quality, stress patterns. The learner as a whole is understood.
u-Learning asks “Where is the Student?” → PALM asks “What is wrong with the Student?”
Three steps to make the AI TA a learning companion that understands you
Your wearable device (such as a Garmin watch) automatically records physiological data such as sleep, heart rate variability and stress levels in daily life, and syncs it to the Uedu platform via Webhook.
The system builds a personal baseline from your data over the past 30 days, using z-score deviation detection to determine whether today's mental and physical state is different from usual, for example 'sleep shorter than usual' or 'stress is higher than usual'.
When you ask AI Teaching Assistant a question, your physiological state is injected into AI's system prompt. AI then adjusts its tone, length and strategy accordingly — for example, slowing down when stress is high, or giving shorter answers when sleep-deprived.
A three-stage pipeline from raw data from wearable devices to AI TA behavioural adaptation
Different wearable devices use different data formats and sampling frequencies. The abstraction layer normalises heterogeneous data from Garmin, Apple HealthKit and Google Health Connect into a unified schema.
Everyone's 'normal' is different — an RMSSD of 40 ms may indicate a good state for Student A, but a stress signal for Student B. The system defines your normal using your own historical data.
PALM's most distinctive design: using natural language rather than numerical rules as the bridge between physiological signals and AI behaviour. LLMs are naturally good at understanding language instructions.
The system has detected that it is late at night, and your stress index today is higher than usual.
I only slept 4.5 hours last night (2 hours less than usual); HRV is low, and Body Battery is at 20%. This information has been injected into the AI context.
Make the responses more concise, with clearer key points and a gentler tone. Rather than saying "you should go to sleep now" directly, naturally show concern for your learning state within the conversation.
Student asks: "Can you help me understand recursion?" — PALM adjusts teaching strategies according to physiological state
Socratic questioning: “What happens when a function calls itself?” Encourages independent exploration and thinking.
direct explanation: "Recursion is when a function calls itself to solve a smaller sub-problem." Clear and concise, with a suggestion to go into more depth later.
step-by-step guidance: "Let's go through it one step at a time." Break it down with concrete examples and an encouraging tone to reduce cognitive load.
Maintain the default mode: sympathetic activation may arise from the physiological response after exercise. After cross-checking the activity data, the system chose not to intervene.
PALM integrates multiple physiological indicators to build a comprehensive picture of physical and mental state
Duration, depth, REM, quality score
Heart rate variability, reflecting the state of the autonomic nervous system
Average all-day stress index
Body Energy Reserve Index
Steps and daily activity
Basic heart rate
Every design decision in PALM places learner wellbeing first
PALM does not send a notification telling you "you should rest". Care is conveyed naturally through the AI TA's conversational tone, answer length and strategic adjustments — you may not notice it, but you will feel the dialogue becoming more attuned to your needs.
Physiological data are used only to adjust AI response strategies in real time and will not be stored in the conversation record. AI will not proactively mention specific data to you unless you ask about it. Data processing complies with research ethics requirements.
Everyone's 'normal' is different. PALM uses your data from the past 30 days to build a personal baseline, and uses statistical deviation (z-score) to judge today's state, rather than a one-size-fits-all absolute threshold.
If there is temporarily no physiological data (for example, if you forget to wear your watch), the AI TA will still work normally. PALM is an additional layer of sensing capability, not a prerequisite for conversation.
PALM’s design must balance technical feasibility and ethical responsibility
Invisible Care is PALM’s greatest strength, and also its greatest ethical challenge. The system “knows” the student’s health status, yet does not let the student notice it. Therefore, transparent informed consent is essential — students must clearly know that physiological data may affect how AI responds.
Learners should have the right to decide whether physiological data affects AI behaviour — even if they have consented to data collection. For example, a Student may consent to provide data for research use, but choose to let the AI TA remain unaffected by it. PALM's modular architecture supports this flexibility.
Everyone's physiological rhythm is different. A Student who sleeps 6 hours each night may perform perfectly normally, but if the system judges by population standards, they may be incorrectly flagged as having 'chronic sleep deprivation'. The personal baseline approach is designed precisely to avoid this problem of defining the individual by the group.
If only students wearing smartwatches could receive better AI support, it would create a two-tier education system. PALM's response strategy: use phone PPG to estimate HRV, and behavioural signals (typing rhythm, response latency) as alternative inputs, ensuring all students can benefit.
A unified health summary layer so users across different devices can all benefit
Automatically sync via Garmin Health API Webhook; supports the Forerunner, Venu, Fenix and other series
Use the Uedu App to read health data from iPhone / Apple Watch
Use the Uedu App to read unified health data from Android devices
The research agenda opened up by PALM, from empirical validation to a fundamental transformation of educational systems
Through a within-subjects crossover design, the same group of Students alternately experience PALM and a standard AI TA, measuring perceived empathy, interaction quality, cognitive engagement and learning outcomes. Key question: do Students notice the adaptiveness? Does physiological monitoring itself change behaviour?
When most Students in a class show increased sympathetic nervous system activation and reduced HRV — indicating collective cognitive overload — the system can prompt the Instructor: "Classroom stress indicators are elevated; consider pausing or switching teaching strategies." PALM extends from an individual teaching assistant to a classroom-level teaching sensor.
In East Asian educational settings, physiological monitoring may be seen as school care; in Nordic contexts, the same system may be felt as surveillance. Multi-site research across at least three cultural contexts (East Asia, Europe and the Americas) is needed to ensure that PALM's design principles have cross-cultural validity.
Within 5 years: consumer EEG headsets and smart glasses with built-in eye tracking will increase cognitive load and attention metrics. Within 10 years: sensors may be embedded in furniture and clothing, allowing learners to use no wearable devices at all. PALM is designed to be scalable, growing as sensing technology advances.
PALM is a core technology of the Uedu Mind subsystem, an application within the PhysioNeuromics dimension of the Educational Omics framework. We explore how physiological data can help AI TAs better understand learners' states, thereby providing more empathetic teaching support.