Inspiration: It is a universal experience for women: walking home alone at night, keys clutched in hand, pretending to be on a phone call just to feel safe. We do this for two reasons: to look busy to strangers, and to feel less lonely.
But a fake phone call is a lie. It can't see the person following you, and it can't comfort you when your heart starts racing.
I wanted to build Andra to bridge that gap. I didn't just want a "security camera" in my pocket; I wanted a companion. I wanted the protective older sister I wish I had on those lonely walks, someone who provides genuine moral support to keep me calm, but who is also wide awake and ready to scream for help if I go silent.
What it does: Andra (Active Neural Defense Response Agent) is a dual-mode AI companion that lives on your laptop/phone. She is the first safety app designed to protect both your mental state and your physical safety.
Sister Mode (Moral Support & Comfort): When you open the app, it looks like a FaceTime call. Andra uses Google Gemini to chat with you naturally. She asks about your day, cracks jokes, and listens to you. This provides critical moral support, lowering your anxiety levels and making you look visibly "connected" to the outside world.
Sentinel Vision (Active Protection): While she chats, she is watching. Using Computer Vision (YOLOv8), she scans the video feed 30 times a second. If a stranger breaches your "Safe Zone" (approaches from behind), she detects them instantly.
The "Silence" Trigger (Dead Man's Switch): This is our core innovation. If Andra detects a threat, she asks, "Do you know this person?" If you are too scared to speak—if you go silent—she doesn't wait. After 8 seconds of silence, she assumes danger, triggers a Red Alert, and uses Twilio to call your emergency contacts automatically.
How we built it: We built Andra using a modular "Sense-Think-Act" architecture in Python:
The Eyes (Vision): We used YOLOv8 and OpenCV for real-time human detection and bounding-box regression to estimate physical distance.
The Brain (Intelligence): We integrated Google Gemini 2.0 Flash. We used advanced prompt engineering to give Andra two distinct personalities: a warm, empathetic listener for "Sister Mode" and a cold, authoritative dispatcher for "Guardian Mode."
The Connection (Dispatch): We used the Twilio API to bridge the digital and physical worlds, allowing our code to place real cellular calls. The UI: To make the experience feel real, we built a custom OpenCV UI engine that mimics the iOS 18 interface, complete with glassmorphism and animated pulse effects.
Challenges we ran into: The "Silence" Problem: Programming an AI to panic when there is no input was incredibly difficult. We had to fine-tune asynchronous timers to distinguish between a "thoughtful pause" in conversation and a "fear-induced silence."
Balancing Emotion & Logic: We struggled to make the AI feel supportive without being distracting. We had to tweak the Gemini prompts to ensure she offered moral support (short, reassuring sentences) rather than long, robotic paragraphs. Latency: Processing video frames and generating AI voice responses simultaneously caused lag. We solved this by implementing multi-threading to handle Vision and Audio on separate cores.
Accomplishments that we're proud of: The "Moral Support" Factor: We didn't just build a tool; we built a vibe. Testing the app feels genuinely comforting, like having a friend in your pocket.
The Dead Man's Switch: We are proud of creating a safety trigger that works even if the user is paralyzed by fear and cannot move. The UI Polish: We went from a black terminal window to a polished, professional "iPhone-style" interface in under 24 hours.
What we learned: Safety is Psychological: We learned that "feeling safe" is just as important as being safe. The moral support aspect of the AI proved to be just as vital as the threat detection.
Multimodal is the Future: Combining Vision (YOLO) and Language (Gemini) creates a context-aware system that is far smarter than standard "panic button" apps.
What's next for Project Andra: Safety is Psychological: We learned that "feeling safe" is just as important as being safe. The moral support aspect of the AI proved to be just as vital as the threat detection.
Multimodal is the Future: Combining Vision (YOLO) and Language (Gemini) creates a context-aware system that is far smarter than standard "panic button" apps.
Log in or sign up for Devpost to join the conversation.