3ayn عين

A real-time AI social awareness assistant for blind and low-vision users — Arabic-first.
Seeing AI tells you what's around you. 3ayn tells you what's happening around you.

Built for the Amazon Industry Program 2026 · AWS Serverless · Java 21 · MediaPipe

---

What it does

Existing accessibility apps are reactive and object-focused: the user taps, the app describes objects. 3ayn is proactive and socially focused — it watches continuously and narrates social events out loud, in Arabic, the moment they happen:

Event	Narration
Someone approaches	شخص يقترب منك
Someone extends a hand	شخص يمد يده نحوك
Someone appears in front of you	شخص أمامك
Crowd / quiet	المكان مزدحم / المكان هادئ
A known person is recognized	أمامك ماريتا، أختك

Plus three on-demand modes: Ask (describe my surroundings), Read (read the text in front of me), and Find (where is my bottle? → "إلى يسارك").

The core architecture decision

A cloud round trip takes 1–3 seconds. "Someone is extending a hand toward you" narrated 3 seconds late is useless — so 3ayn splits by latency requirement:

 LATENCY-CRITICAL (continuous)          LATENCY-TOLERANT (user-initiated)
 ─────────────────────────────          ──────────────────────────────────
 MediaPipe pose estimation               Ask   →  Bedrock Nova 2 Lite
 + geometric heuristics                  Read  →  Textract
 runs IN THE BROWSER                     Find  →  Rekognition Labels
 zero network calls                      Who   →  Rekognition Faces
 instant narration                       Speak →  Polly (Arabic neural voice)

Real-time narration never waits for a network. Everything the user explicitly asks for runs on AWS, where managed AI earns its latency.

AWS services used

Bedrock (Nova 2 Lite) scene description · Textract OCR · Rekognition object labels + face matching · Polly Arabic TTS · Lambda (7 functions, Java 21) · API Gateway · DynamoDB (users + face profiles) · SAM infrastructure-as-code · Amplify hosting

Repository structure

3ayn/
├── files/
│   └── 3ayn-app.html          ← the entire frontend (single file)
├── 3ayn-backend/
│   ├── template.yaml          ← the whole AWS stack, declared
│   ├── samconfig.toml
│   └── ThreeAynFunction/      ← Java 21 handlers (7 endpoints)
└── docs/                      ← technical report, screenshots

Run the frontend

cd files
python -m http.server 8000
# open http://localhost:8000/3ayn-app.html  (camera requires localhost or HTTPS)

Register with your name, then paste the backend URL in Settings → Test connection.

Deploy the backend

cd 3ayn-backend
sam build
sam deploy --guided        # stack: threeayn, region: eu-west-2

Copy the ApiBaseUrl output into the app's Settings. Full endpoint contracts, smoke tests, and cost notes: 3ayn-backend/README.md.

API

Endpoint	Service	Returns
POST /ask	Bedrock Nova 2 Lite	Arabic/English scene description
POST /read	Textract	Text in frame, read aloud
POST /find	Rekognition Labels	Object + direction (left/right/front)
POST /who	Rekognition Faces	Name + relation of enrolled person
POST /enroll	Rekognition + DynamoDB	Caregiver enrolls a known face
POST /user, GET /user/{id}	DynamoDB	Wearer profile + spoken welcome
POST /speak	Polly	Base64 MP3, Arabic neural voice

Privacy & consent

Face recognition matches only people deliberately enrolled by a caregiver, with their knowledge. Strangers are never stored, fingerprinted, or clustered. No authentication by design — 3ayn is a single-wearer assistive device; a profile personalizes narration, it is not an account.

Roadmap

Event history per wearer → caregiver dashboard (live log, alerts) · "someone is speaking to you" detection · voice-command language switching (Transcribe) · smart cane companion (ESP32 + ultrasonic) · smart glasses form factor.

Team

Built by Team 8 — Amazon Industry Program 2026 (IEEE / Women in Engineering, AUB).