EMind

The system detects the user’s facial expression and identifies a “tired” emotional state.
The system detects the user’s facial expression and identifies a “neutral” emotional state.

Inspiration

People with ADHD often struggle to regulate attention and emotion in real time, especially when switching between tasks or facing tight deadlines. Traditional tools like to-do lists, reminders, or focus timers are reactive — they require intentional effort and don’t adapt to a user’s moment-to-moment state. We wanted to create a proactive, emotion-aware assistant that could detect subtle behavioral and emotional cues and respond in a supportive, personalized way — just like a calm, observant friend who notices when you’re drifting, overwhelmed, or stuck.

What it does

EMind is a privacy-first AI assistant that passively senses signals from:

👁️ Facial expressions
⌨️ User behavior (typing, tab-switching, screen interaction)
🔊 Ambient sound

Using these signals, it infers one of several ADHD-relevant states: Focused, Drifting, Overwhelmed, Stuck, Hyperfocus, or Transition Fatigue.

Once a state is inferred, EMind uses an LLM to generate 1–2 sentence context-aware prompts — for example: “You seem overwhelmed. How about a short stretch break before diving back in?”

How we built it

DeepFace → for facial emotion and attribute recognition using pre-trained CNNs (TensorFlow/Keras backend).
OpenCV (cv2) → to handle live video capture, frame extraction, and image preprocessing.
Pillow (PIL) → for decoding and manipulating image arrays.
NumPy → for efficient numerical computation and vector operations.
TensorFlow (CPU) → to run emotion detection and classification models locally.
OpenAI API (LLM) → to convert classifier outputs into natural-language, empathetic feedback.

What we learned

Integrate computer vision and LLM-based reasoning in real-time.
Translate raw emotion classification into human-centered interventions.
Design systems that respect user privacy and mental well-being.
Apply principles from affective computing and behavioral psychology to product design.

Challenges we ran into

Cross-modal fusion — combining visual, auditory, and behavioral inputs into a unified interpretation required careful tuning.

Accomplishments that we're proud of

Building a device that will really help people with ADHD to focus and work independently.
Successfully used DeepFace and TensorFlow to detect subtle emotional shifts like frustration, confusion, or hyperfocus from live webcam input.
Designed a privacy-first system that processes signals locally, ensuring sensitive emotional data never leaves the user’s device.
Created a lightweight, human-like response system powered by an LLM that translates raw classifier output into empathetic, contextual prompts.
Developed a Tkinter popup interface that delivers state-aware interventions (e.g., suggesting music, stretching, or short breaks) without breaking focus.
Bridged psychology and AI, applying principles from affective computing and ADHD research to create a tool that genuinely supports emotional regulation and focus.

What's next for EMind

Platform expansion — build a mobile and browser-based version that runs quietly in the background while users work or study.
Collaboration with ADHD communities — partner with clinicians and support groups to validate EMind’s effectiveness in real-world environments.