Offline AI tool for Parkinson’s rigidity assessment using Gemma 4, it is built for low-resource healthcare settings.
# 🧠 RigidityIQ
### Offline Parkinson's Rigidity Assessment for the Last Mile
> ⚡ Runs fully offline. Clinically grounded. Built for deployment where neurologists don't exist.
**Track:** Health & Sciences · **Model:** Gemma 4 E2B via Ollama
**Stack:** Python · Gradio · ChromaDB · SQLite · sentence-transformers
---
## The Problem Is Not Knowledge. It's Reach.
I spent a year on this problem the wrong way.
I trained an ordinal regression pipeline on 147 patients from the WearGait-PD dataset,
93 synchronized IMU sensor channels, 14 body locations, strict subject-independent evaluation.
Some folds achieved a Quadratic Weighted Kappa of 0.53. The model worked.
Then I asked the question that should have come first: **who can actually use this?**
Running it requires 14 wearable sensors, 100Hz synchronization, and specialist inference hardware.
That is a research lab. Not a district hospital in northern Uganda, where internet is unreliable
and there is fewer than one neurologist per million people.
Uganda has approximately **6–7 trained neurologists for 47 million people**,
0.03 per 100,000, versus 2.96 in high-income countries. Patients with Parkinson's disease
go unassessed for years. Not because the clinical knowledge doesn't exist.
Because the tools to apply it never reach them.
RigidityIQ is my answer to that gap, it is not built for the lab, but for the last mile.
It assesses rigidity, one of the three cardinal motor symptoms of Parkinson's disease. Rigidity is an abnormal increase in muscle tone that stiffens the limbs, shortens the stride, and kills the natural arm swing that most of us never think about. A community health worker (CHW) can observe it. Until now, they had no clinical tool to grade it.
RigidityIQ is a support tool.
---
## Demo
---
## What RigidityIQ Does
A Community Health Worker (CHW) enters structured clinical observations —
walking speed, arm swing, posture, muscle resistance, free-text notes.
RigidityIQ returns:
- A * …