Cost wall
Specialist wearables often cost thousands. Guide dogs involve long waits. eyEar targets under $50 so independence is not a privilege.
Wearable assistive AI · prototype in testing
What if your ears could see?
eyEar is a wearable that clips onto your shirt, belt, or cane and describes what's around you—obstacles, signs, food, people—through a small speaker. We're building it to cost under $50.
- Working prototype
- Under $50 target
- Hands-free
- eyEar Cortex
Product
eyEar, in one breath
A clip-on with a camera and proximity sensor, paired with your phone. It picks up obstacles, reads signs, identifies objects, and describes the scene around you—all through a small speaker, hands-free.
What makes it different is eyEar Cortex—the layer that shapes how the device responds based on what you're actually doing, not just what it sees.
See it run
Prototype demo
Four demos in one video—real hardware, real flows. Opens on YouTube.
Why we exist
Most options cost thousands, need your hands, or just list what’s there without helping you act on it
Over a billion people live with meaningful vision loss. For a lot of them, the problem isn’t that tools don’t exist—it’s that they’re too expensive, don’t fit how people actually move through the world, or just aren’t helpful enough.
Wrong shape
Phone apps tie up your hands. One-size eyewear ignores cane-first users, guide-dog partners, and people who simply will not wear glasses for assistive tech.
Labels, not next steps
“Chair, table, person” is not enough. People need steps, clock positions, social context— the kind of answer a thoughtful human would give, at the speed of software.
44% vs 79%
Employment: blind or visually impaired vs. without disabilities (U.S.)
AFB statistics$2k–$4.5k+
Typical vision-aid price bands vs. eyEar’s sub-$50 goal
Cortex
Scenario-tuned responses—built to say what to do, not only what is there.
Our journey
Where we've been and where we're headed
Nine months of work, starting with conversations and ending with a device that runs.
-
Done
Insight & listening
Deep conversations on daily friction—navigation, meals, shopping, social cues—before writing a line of product code.
-
Done
Design partners
Ongoing blind co-designers set priorities: what is actionable vs. noise, and how audio vs. haptics should balance.
-
Done
Prototype v1 + Cortex seed
First wearable loop: sensors, vision API, and early eyEar Cortex prompts shaped by real scenarios.
-
Now
Working prototype & active testing
We are here: obstacle detection and scene description are working. Haptic feedback is in. Cortex scenarios are being expanded. The device needs more testing before it's ready to hand to users.
-
Next
Pilot programs
Structured pilots with schools, community orgs, or clinics—measuring independence gains and refining onboarding.
-
Then
Manufacturing & compliance path
Design for production, supply chain, labeling, and the right regulatory posture for the markets we enter.
-
Ahead
Scale & Cortex library growth
Expand the scenario library to hundreds of modes—every new feature ships through the same Cortex pattern you see below.
What we have built
A tested wearable, not a slide deck
Today’s stack: compact ESP32-S3, camera, ultrasonic sensor, discreet clip-on (belt, shirt, cane, or neck) with a phone for hands-free cloud vision when you want it.
- Ultrasonic safety: fixed distance thresholds trigger obstacle alerts before vision inference does.
- Vision + language: Gemini 2.0 Flash for scene understanding and description.
- Haptic feedback: vibration cues for obstacle alerts—no audio required.
- eyEar Cortex: the scenario layer that shapes responses based on context, not just what the camera sees.
Roadmap sections describe direction; not every layer is in every build yet.
The USP
eyEar Cortex — scenario intelligence, not a chatbot bolt-on
Off-the-shelf vision models describe pictures. eyEar Cortex is the product: a growing library of situations, prompts, and response shapes co-designed with blind partners so the device speaks like a skilled orientation partner—steps ahead, clock positions, social awareness—not a firehose of object labels.
Every future feature (there will be hundreds) lands as a Cortex scenario: a defined context, safety rules, output format, and test cases. Tap any row below to see how we document what “good” sounds like. This pattern scales as we add modes.
Navigation & transit
Empty seat on bus or train
Goal: Find sittable space and approach path.
Example output: “Empty seat two steps ahead, slightly to your right; aisle clear.”
Crosswalk & curb alignment
Goal: Align to crossing box and confirm signal phase when visible.
Example output: “Curb straight ahead; crosswalk stripes underfoot; pedestrian signal on your left shows walk.”
Indoor corridor & doorway
Goal: Center in hall, announce doors and turns.
Example output: “Hall continues 8 meters; open doorway 2 o’clock—likely restroom sign.”
Elevator floor & button panel
Goal: Read floor indicator and describe control layout.
Example output: “Panel on right; ‘3’ illuminated; braille strip along bottom edge.”
Dining & kitchen
Plate layout by clock position
Goal: Map food for utensil approach.
Example output: “Avocado at 3, sandwich at 6, jam at 9.”
Cafeteria line & counters
Goal: Describe stations and shortest queue hint.
Example output: “Three stations: salad left, hot entree center, cashier right; middle line shortest.”
Stove & appliance state (when visible)
Goal: Surface obvious on/off cues—always paired with ultrasonic safety elsewhere.
Example output: “Front-left burner glow red; kettle on rear right.”
Social & people
Who is in the room & group drift
Goal: Orientation to faces, proximity, whether attention shifted.
Example output: “Three people in arc ahead; two turned toward the door—conversation may have moved.”
Pick clothing from a row
Goal: Distinguish shirts by color/pattern and position.
Example output: “Blue shirt, second from the left on the rack.”
Find a familiar face in a crowd
Goal: Directional cue + distance when partner provided reference.
Example output: “Possible match 11 o’clock, ~4 meters, navy jacket.”
Reading, docs & learning
Physical book paragraph flow
Goal: Page boundaries, read order, bookmarking (roadmap).
Example output: “New paragraph starting; header ‘Chapter 4’ at top of page.”
Handwritten note Q&A
Goal: Semantic search over captured text—“what date is mentioned?”
Example output: “Date line: March 15 near signature.”
Slide or whiteboard overview
Goal: Bullet structure for classroom or meeting access.
Example output: “Three bullets: Budget, Timeline, Risks—timeline bullet has a sub-list.”
Safety & hazards
Head-level obstacles (with vision + ultrasonic)
Goal: Complement cane sweep with branch, sign, or cabinet context.
Example output: “Low branch ahead at forehead height; step left to clear.”
Escalator & stair lip
Goal: Clear edge and direction of travel cues.
Example output: “Escalator mouth 2 steps ahead; handrail on your right.”
Construction & temporary barriers
Goal: Narrate tape, cones, and alternate path hints.
Example output: “Yellow tape crosses sidewalk; gap at 10 o’clock.”
New scenarios ship continuously—same Cortex pipeline: capture → scenario detect → guided prompt → spoken (or haptic) response. Partner testing decides what graduates to default modes vs. optional packs.
Architecture
How it works
Camera + ultrasonic in parallel. Description runs through Cortex and a vision model; obstacle alerts stay on deterministic sensors.
Camera
Scene frames
eyEar Cortex
Scenarios · prompts · output shape
Vision model
e.g. Gemini 2.0 Flash
Voice out
Spoken guidance
Ultrasonic
Distance thresholds · alert path
Collision safety uses fixed distance rules, not vision inference—by design.
Seeed XIAO ESP32-S3 · Gemini 2.0 Flash · Python · working prototype
Platform roadmap
One core, many forms, blind-first modes
A modular core for eyewear, cane, neck, or belt—plus purpose-built experiences and a layered safety story when connectivity drops.
Multi–form factor
One heart, many wear styles—cane users, guide dog partners, eyewear-averse users included.
Seven experience lanes
Scene, plate, books, documents, expression, wayfinding, hazards—each a Cortex-heavy mode.
Offline-first safety
Physics-first sensors, optional on-device fallback, rich cloud when connected.
Institutions
Schools, NGOs, rehab—alongside consumers—for reach at scale.
Roadmap experience names
- Scene Narrator
- Danger Spotter
- Plate Analyzer
- Book Reader
- Expression Reader
- Document Scanner
- Wayfinder
Who we serve
Blind and low-vision people first—especially those priced out today
Students, working-age adults, and older adults with progressive loss. Under $50 remains the north star.
- Independent navigation with obstacle awareness
- Situational awareness through Cortex-tuned descriptions
- Social inclusion: who is present, who approaches, group dynamics
Ethics & safety
Responsible AI
- Privacy. Images go to inference APIs; not retained by eyEar. Bystander capture is disclosed.
- Bias. We test across contexts and refine Cortex when outputs miss—ongoing, not one-time.
- Complement. Works with canes, dogs, and orientation skills—not a replacement.
About
Team
Sia Gupta
Inventor & Lead Engineer
School: Tesla STEM High School
Location: Redmond, WA
Grade: 11th
Atishay
Research & Human-centered Design
School: North Creek High School
Grade: 10th
Collaborate
Partner with us
We're two high school students building something we think matters. We're at the stage where feedback and connections to the right people would make a real difference.
We're especially interested in:
- Organizations that can connect us with blind and low-vision individuals for structured testing and feedback.
- Experts in accessible design who can help refine hardware, software, and audio experiences.
- Communities and schools that want to help build something that measurably improves independence and confidence.
If that sounds like you or your organization, reach out and let us know how you'd like to help—we read every message.