Smart Glasses · Computer Vision · Memory

👓 Smart Glasses
Memory Assistant

Wearable memory for the moments when your brain needs context faster than you can pretend you remember. Press a button, hear who you're looking at.

8ms
ArcFace inference · CPU
99.7%
LFW benchmark accuracy
<2s
button press → spoken recall

The Problem

Face recognition as a medical accommodation has been a solved problem in research labs for years. The solutions that have reached actual users are almost all awkward: dedicated apps that require pulling out your phone mid-conversation, systems with significant cloud latency, or experimental hardware that announces itself visually.

For sales professionals and enterprise networkers, the problem is different but structurally similar. CRM systems capture data after the fact, require manual entry, and sit behind a phone screen you can't easily check while maintaining eye contact.

A system that whispers context into your ear as you approach someone is categorically different from one that requires you to pause the conversation to look something up.

What It Does

XKOENPRO smart glasses (12MP cam) stream video to a paired laptop
MediaPipe FaceMesh — 478-landmark detection, locates and normalises the face
InsightFace ArcFace — 512-dimensional embedding, cosine match in <8ms on CPU
SQLite WAL database — persons, sightings, meeting notes, transcript segments
Whisper ASR + pyannote diarization — who said what in every conversation
Claude recall API — natural language response from structured memory data
AVRCP button → TTS → Bluetooth speaker — heard only by the wearer

Who It's For

Prosopagnosia (face blindness)

~2.5% of the population. Constant exhausting social anxiety. A reliable, private, wearable system is a meaningful accommodation — not a novelty.

Sales & account management

Hundreds of relationships across events, calls, and meetings. "I remember you mentioned your daughter was starting university" is worth more than a business card.

Aged care & dementia support

A system that introduces a visitor by name and relationship context before the patient has to ask. Measurable patient outcome value.

Conference & event professionals

Badge-free recognition at private summits. Premium networking infrastructure that nobody else has.

Technical Capabilities Today

CapabilityStatusDetail
Face recognition✓ LiveArcFace 512-d, 99.7% LFW, 8ms CPU
Multi-face tracking✓ LiveUp to 10 simultaneous faces, 30fps
Conversation capture✓ LiveWhisper + pyannote, local
Memory database✓ LiveSQLite WAL, unlimited persons
Button recall✓ LiveAVRCP → Claude → TTS → speaker
Cloud sync→ AWSDynamoDB + Lambda
Multi-device→ AWSRequires cloud memory layer
Wireless glassesRoadmapPhone bridge in progress
What AWS infrastructure unlocks

DynamoDB — Multi-device memory sync. Recall works when the laptop is off. Sub-100ms lookup. Encrypted per-user partitions.

Transcribe Streaming — 200ms ASR latency, built-in speaker diarization. Replaces local Whisper for real-time capture.

Bedrock (Claude) — Cloud AI inference, no local model required. IAM auth, audit logs, no API key exposure.

Lambda — Recall API when the laptop is off. Callable from a phone app. Scales to B2B deployment.

"A cognitive prosthetic for social memory. The only system that passively builds social context in real time, speaks recall aloud, and runs fully offline on commodity hardware."
GTM strategy → live demo → ← back to showcase