Case study · Summer 2025

Real-Time ASL Translator

An offline-first multimodal pipeline translating webcam ASL gestures into English text and speech.

Discipline
Multimodal AI · Accessibility
Technology
OpenCV · Gemma 3n · FastAPI · Ollama
CAMERA INPUT
GESTURE → TEXT → SPEECHOFFLINE · PRIVATE

I combined OpenCV input processing with Gemma 3n inference behind a FastAPI service, then deployed the pipeline locally through Ollama for privacy and low-connectivity use.

Live webcamInput
Text + speechOutput
Offline firstDeployment
AccessibilityFocus

How the pieces connect

01Live webcam
02OpenCV · Gemma 3n
03Text + speech

What made the problem worth solving

A real-time accessibility tool has to connect perception, language, and presentation while respecting the privacy and connectivity constraints of the people it is meant to support.

Turning the problem into an engineering plan

  1. 01

    Captured and prepared live gesture input through OpenCV.

  2. 02

    Used Gemma 3n to connect multimodal observations to English output.

  3. 03

    Ran the system locally via Ollama to avoid sending sensitive camera input to a cloud service.

The prototype connected multimodal AI with a concrete human need and pushed me toward local-first systems where privacy, responsiveness, and access matter.

Next case studyYouTube RAG