Parxx

Login/Signup Page
Dashboard
Real Time Camera Page
Immediate Report of Symptoms
Summary of Each Report
Sending Email to the caregiver
Generated PDF Report
AI Voice Assistant

Inspiration

Parkinson’s disease is progressive, with symptoms that gradually change and worsen over time. For individuals already diagnosed, tracking mobility is critical for adjusting treatment, therapy, and daily care. At the same time, many people begin showing subtle motor changes, such as slower movement, reduced arm swing, long before a formal diagnosis. These early signs are often missed in everyday settings until noticeable decline has already occurred.

We were inspired to build a solution that supports both early detection and continuous monitoring. Parxx helps identify early warning signs in at-risk individuals while enabling caregivers and clinicians to objectively track disease progression over time. Our goal is to shift Parkinson’s care from occasional observation to proactive, data-driven insight.

What It Does

Parxx is an AI-powered movement monitoring system designed to detect early signs of Parkinson ’s-related motor changes and track disease progression over time.

The system works as an end-to-end pipeline that transforms video into actionable clinical insights:

Extracts body movement from video using pose estimation to detect and track skeletal keypoints
Analyzes motion over time to measure clinically relevant mobility features, including:
- Stride consistency and gait stability
- Posture alignment and changes
- Movement symmetry and amplitude
- Freezing episodes and fall detection
Applies a severity scoring model (0–10 scale) to classify events such as:
- Parkinson-like gait patterns
- Freezing episodes
- Fall incidents
Builds a personalized monitoring history and tracks trends over time
Identifies risk patterns or functional decline using anomaly detection and structured thresholds
Generates automated clinical-style reports summarizing mobility observations and behavioral changes
Displays insights on a web dashboard with historical tracking and visual indicators
Sends real-time email alerts when high-severity events are detected
Provides an AI-powered virtual assistant that explains reports in plain language and highlights key findings

Parxx transforms passive video into proactive, continuous insight, enabling earlier awareness and more informed care decisions.

How We Built It

Parxx was built as a full-stack AI monitoring pipeline.

1. Movement Extraction

We used pose estimation to extract skeletal keypoints from video frames in real time. This allowed us to analyze joint positions, movement symmetry, velocity changes, and posture stability over time.

2. Motion Feature Engineering

From the extracted pose data, we derived clinically relevant signals such as:

Reduced arm swing symmetry
Freezing patterns (prolonged immobility)
Fall detection based on posture and movement changes

These features were analyzed over time to capture meaningful behavioral shifts.

3. Severity Scoring System

We implemented a structured severity model (0–10 scale) that classifies events into:

Mild motor irregularity
Freezing episodes
Fall events

This severity system drives alerts, report generation, and dashboard prioritization.

4. Backend & Data Pipeline

We built a backend API that:

Processes inference results
Stores session events in Firestore
Generates structured AI summaries
Creates downloadable PDF reports
Sends emails to the caregiver

5. AI Narrative Generation

We integrated an AI model to transform structured movement data into clinical-style summaries that include:

A session overview
Key observations
Interpretation of findings
Suggested next steps
Safety considerations

The narrative avoids diagnostic claims while providing meaningful context for caregivers.

6. Dashboard & Alerts

The frontend dashboard was built to:

Display real-time session events
Show severity indicators visually
Store historical reports
Provide downloadable PDFs

We also implemented automated email alerts for high-severity events such as freezing or falls.

Challenges We Ran Into

Distinguishing normal variation from true motor decline — Human movement is inherently noisy, requiring careful threshold tuning.
Avoiding false alarms — Fall and freeze detection needed balance between sensitivity and reliability.
Parsing AI-generated structured output cleanly — We initially displayed raw JSON before implementing normalization.
Persisting real-time session data properly — Early versions lost data on refresh before Firestore integration.
Balancing medical framing responsibly — We carefully structured language to provide insights without implying diagnosis.

Accomplishments That We're Proud Of

Building a complete end-to-end AI pipeline in a short timeframe
Designing a practical severity scoring system tied to actionable alerts
Generating structured, professional PDF reports automatically
Integrating real-time detection with historical tracking
Delivering a system that feels like a clinical monitoring tool rather than a prototype

Most importantly, we demonstrated how computer vision and AI can meaningfully support long-term neurological care.

What We Learned

Healthcare AI requires thoughtful thresholds, not just model output
Data persistence and auditability are essential for trust
Clear communication matters as much as technical accuracy
Even rule-based severity systems can be powerful when paired with strong UX
Responsible framing is critical in health-related technologies

What’s Next for Parxx

Personalized baseline calibration for each individual
Long-term trend visualization across weeks and months
Clinician-facing analytics dashboard
Secure cloud deployment with encrypted storage
Integration with wearable sensor data
Clinical validation partnerships

Our long-term vision is to make Parxx a continuous monitoring companion — helping shift Parkinson’s care from reactive treatment to proactive insight.