Labmino

AI
computer vision
embedded systems

We start with the people closest to the problem, then build and test with them.

Explore our work or meet the team

Samsung Solve for Tomorrow · 2026: Global Ambassador
Samsung Solve for Tomorrow Global · 2025: Top 20 Worldwide
Samsung Solve for Tomorrow Indonesia · 2025: 1st Place + People’s Choice

flagship project

Active · RunSight · since 2025

AI glasses for
visually impaired runners.

“RunSight began with Dimas, a friend with cataract blindness who kept showing up in sport.”

RunSight reads the track and the people ahead, then gives spoken guidance to visually impaired runners in real time.

AIComputer VisionEmbedded SystemsAccessibility

Open full project Visit runsight.labmino.com

in the field

Built to perform in the real world.

84.23%

Lane segmentation IoU

88.24%

Person detection IoU

>10 FPS

Real-time target

Offline

No internet required

capabilities

What we do best.

Research

We listen first.

Code comes later. We start by spending time with the people a product is meant to help, because the real problem is usually more specific than the first brief.

Full stack

We connect every layer.

Model, device, app, backend. The same small team works across the stack, so tradeoffs stay visible and nothing gets handed off too early.

Prototyping

We test outside the room.

A lab demo can hide too much. We put prototypes into the setting they were built for, watch what breaks, and change the work from there.

Accessibility

We design where mistakes matter.

When a product works for someone with less room for error, it usually becomes clearer and more reliable for everyone else too.

process

How the work actually happens.

The useful insight came from a conversation. At first, we thought the problem was general navigation. Field interviews made it sharper. Running on a public track is not the same as walking through a street.

Being wrong early is part of the work. Our first model could not keep up with changing track conditions. We stopped trying to patch it and rebuilt the pipeline around several smaller models working together.

A quiet system can be safer than a confident one. RunSight speaks to someone moving at speed. If the system is not sure, staying silent is better than giving a risky cue. That rule shapes the model thresholds, smoothing, and voice guidance.

the team