Context-aware AI companion · Silent-danger-first safety · GPS navigation with transit · Natural conversation
Turn on → it works. Say where → it navigates. Say nothing → it keeps you safe.
"The biggest daily challenge is taking the bus." — SAVH Advocate
1.3 billion people live with vision impairment. Existing AI glasses describe scenes — but they don't navigate. They can't guide a blind person to a bus stop, tell them which bus is arriving, or help them board the right one. Asirive Cortex solves this by providing true independent mobility through a highly advanced, <$150 chest-mounted wearable.
Powered by Gemini 3.1 Flash Live, Cortex doesn't just have one personality. It dynamically switches between 6 behavioral profiles based on context:
- IDLE: Silent and observant. Only speaks for overhead hazards.
- OUTDOOR_NAV: Turn-by-turn GPS companion.
- INDOOR_NAV: Proactive camera guidance when GPS is lost.
- BUS_WATCH: Laser-focused on reading bus numbers and LTA DataMall arrivals.
- TRANSIT: Quietly announces stops and remaining journey time.
- EXPLORE: Detailed scene narration on demand.
Traditional assistive devices spam the user with alerts about everything (people, dogs, cars). Asirive Cortex filters out what you can naturally hear. Using local YOLO + Depth sensing, it warns ONLY about silent dangers:
- 🧱 Walls, poles, and overhead obstacles
- 🕳️ Stairs, curbs, and drop-offs
- 🚗 Approaching vehicles from blind spots
- Feedback escalates from voice alerts to haptic pulses as distance closes. (<100ms latency, 100% Offline)
Multi-leg routing made easy: Walk → Bus → MRT → Walk. Features voice navigation, real-time stop counting, and precise arrival detection.
Ask anything: "What do you see?", "Read that sign", "Where did I leave my keys?". Built-in local SQLite and Supabase cloud sync allows Cortex to remember objects and locations for you.
Asirive Cortex has been practically designed and tested with the Singapore Association for the Visually Handicapped (SAVH). Our live demonstration features four core stations proving real-world viability:
| Station | Focus | What It Tests |
|---|---|---|
| 1️⃣ Indoor Safety | Obstacle Avoidance | Tiered safety protocols: voice alerts escalating to haptic pulses for silent indoor hazards. |
| 2️⃣ Ask AI | Scene Understanding | Real-time multimodal Q&A with Gemini Live and function calling. |
| 3️⃣ Outdoor Nav | Waypoint Tracking | Turn-by-turn voice guidance and spatial awareness outdoors. |
| 4️⃣ Bus Arrival | Public Transit | LTA DataMall integration combined with live YOLO bus detection to identify arriving buses. |
Cortex operates on a 5-Layer "Brain" architecture, balancing lightning-fast offline reflexes with deep cloud intelligence.
Click to view the 5-Layer AI Brain details
| Layer | Name | Role | Tech Stack | Latency | Device |
|---|---|---|---|---|---|
| L0 | Guardian | Safety-critical detection + haptic alerts | YOLO v11 NCNN + GPIO | <100ms | RPi5 (Offline) |
| L1 | Learner | Adaptive open-vocabulary detection | YOLOE | ~200ms | RPi5 |
| L2 | Thinker | Scene understanding, reading, Q&A | Gemini 3.1 Flash Live | ~500ms | Cloud |
| L3 | Guide | Intent routing + GPS + transit | Fuzzy router + LTA DataMall | <5ms | RPi5 |
| L4 | Memory | Object recall, cloud sync | SQLite + Supabase | ~1ms | Hybrid |
graph TD;
User[User Audio/Video] --> VAD[Silero VAD];
VAD --> STT[Whisper/Cartesia STT];
STT --> Router[Intent Router 97.7% Acc.];
Router --> L0[L0/L1 Guardian: YOLO <100ms];
Router --> L2[L2 Thinker: Gemini Vision];
Router --> L3[L3 Guide: Navigation & GPS];
L0 --> Haptic[Haptic Pulse / Voice Alert];
L2 --> Convo[Conversation / Q&A];
L3 --> Nav[Voice Guidance];
(Note: The Raspberry Pi 5 runs fully standalone. The optional Laptop Dashboard is purely for monitoring/dev).
All safety-critical features rely on open-ear or bone conduction earbuds to ensure the user's natural hearing is never obstructed.
| Component | Purpose | Cost (Est.) |
|---|---|---|
| Raspberry Pi 5 (4GB) | Core compute module | $60 |
| Camera Module 3 Wide | 1080p @ 30fps scene capture | $35 |
| NEO-6M GPS & BNO055 IMU | Positioning & Heading | $20 |
| Vibration Motor & Button | Haptic alerts & input control | $3 |
| USB Lavalier Mic | 16kHz voice input | $8 |
| Open-Ear Bluetooth Earbuds | Safe audio feedback | $20 |
| 5000mAh Power Bank | ~4 hours active runtime | $10 |
| TOTAL | **~ $156** |
- Hardware: RPi5 (4GB), Camera Module 3 Wide, Open-ear Bluetooth earbuds.
- API Keys: Gemini API, LTA DataMall, Google Maps.
- Python: 3.11+
git clone https://github.com/IRSPlays/ProjectCortex.git
cd ProjectCortex
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt
# On RPi5:
sudo apt install python3-picamera2 espeak-ngCreate a .env file in the root directory:
GEMINI_API_KEY=your_gemini_key_here
LTA_ACCOUNT_KEY=your_lta_key_here
GOOGLE_MAPS_API_KEY=your_maps_key_here
SUPABASE_URL=https://your-project.supabase.co
SUPABASE_KEY=your_supabase_anon_key# Production mode (Standalone RPi5)
python rpi5/main.py
# Optional: Run the monitoring dashboard on a laptop
python laptop/gui/cortex_ui.pyCurrent Status: V2.5 (SAVH demo polish, hazard cooldown fixes). Next Up (V3.0): Custom PCB, integrated audio, 57% lighter waterproof enclosure.
Asirive Cortex is licensed under the GNU General Public License v3.0 or later. See LICENSE.
Built for independence. Powered by Gemini. Designed with SAVH.
© 2026 Asirive. Built by Haziq, founder of Asirive.
