Project Info

SafeLine

Inspiration

Domestic abuse affects millions of people worldwide, and many victims are unable to safely seek help. Fear, coercion, and constant monitoring often prevent them from accessing resources, learning about their rights, or reaching out to others. In many cases, even searching for support online is a risk that can likely lead to danger. We were inspired by the reality that the people who most need help often have the least safe access to it. We wanted to design a solution built specifically for those constraints. It needed to be something discreet, protective, and aware of the complexities of abusive situations. SafeLine was created with one central goal: to support victims in ways that are practicable, subtle, and centered entirely on their safety.

What it does

SafeLine is a covert safety platform disguised as a standard notes app. By creating a note with a specific password as the title, users unlock a secure system designed for documenting and responding to domestic abuse. Once accessed, the app becomes a protected space for recording incidents through text, audio, and photos while maintaining discretion and privacy. The platform has 5 main features: Real-time abuse detection is powered by on-device voice activity detection and lightweight threat classification. When harmful or escalating language is detected, the app automatically logs the interaction, generates a structured summary, and securely stores it as evidence. Real-time abuse detection is powered by on-device voice activity detection and lightweight threat classification. When harmful or escalating language is detected, the app automatically logs the interaction, generates a structured summary, and securely stores it as evidence. Emergency escalation allows victims, when they cannot safely place a call themselves, to trigger a notification that provides authorities with critical information such as name, address, age, and a summary of recent incidents. Emergency escalation allows victims, when they cannot safely place a call themselves, to trigger a notification that provides authorities with critical information such as name, address, age, and a summary of recent incidents. A reviewer dashboard allows trusted advocates to monitor multiple users, review incident histories, receive imminent danger alerts within seconds, and determine whether escalation to law enforcement is necessary. A reviewer dashboard allows trusted advocates to monitor multiple users, review incident histories, receive imminent danger alerts within seconds, and determine whether escalation to law enforcement is necessary. An integrated AI assistant provides guidance on legal rights, safety planning, medical concerns, and situational advice using the user’s documented history as context. An integrated AI assistant provides guidance on legal rights, safety planning, medical concerns, and situational advice using the user’s documented history as context. A protected communication system enables safe chats with trusted reviewers, experts, lawyers, friends, and support contacts without exposing activity. A protected communication system enables safe chats with trusted reviewers, experts, lawyers, friends, and support contacts without exposing activity. Additional features Automatic reversion to safe mode ensures the app returns to its normal notes interface if left unattended (using facial recognition), preventing accidental exposure. Automatic reversion to safe mode ensures the app returns to its normal notes interface if left unattended (using facial recognition), preventing accidental exposure. Secure documentation allows users to manually add incidents. All data is encrypted and can optionally be shared with a trusted independent reviewer for oversight and support. Secure documentation allows users to manually add incidents. All data is encrypted and can optionally be shared with a trusted independent reviewer for oversight and support.

How we built it

📱 Frontend (Native iOS Experience): Built fully in SwiftUI with an MVVM architecture, delivering a fast, seamless app that feels indistinguishable from a normal everyday Notes-style interface. 🔒 Privacy + Secure Access Control: We use CryptoKit passcode hashing and Apple Vision face detection to automatically lock the app when someone else is nearby, protecting users instantly and silently. 🎙️ Real-Time On-Device Threat Monitoring: Powered by AVAudioEngine and Apple Speech Recognition, the app can detect and log suspicious situations as they happen, without requiring the user to manually activate anything. 🤖 AI Reasoning + User Support: Integrated with OpenAI’s multimodal models, our system can classify incidents from transcripts and provide contextual, expert-informed guidance in real time. ☁️ Backend Infrastructure + Persistent Chat Storage: A dedicated web server backed by Supabase (Postgres + Auth) securely stores conversations and enables reliable syncing and reviewer communication. 📞 Autonomous Emergency Escalation: Using Twilio, the platform can automatically trigger phone calls and alerts during high-risk moments, ensuring help is contacted even when the user cannot.

Challenges we ran into

Privacy-Preserving Detection: Designing an edge-triggered audio system like Siri using RMS thresholds and semantic threat analysis without continuous recording, while carefully tuning models to reduce false positives and still capture the majority of abusive incidents. Privacy-Preserving Detection: Designing an edge-triggered audio system like Siri using RMS thresholds and semantic threat analysis without continuous recording, while carefully tuning models to reduce false positives and still capture the majority of abusive incidents. Secure, Low-Latency Alerting: Building an encrypted server pipelines that delivers emergency notifications in seconds, balancing rapid response times with strict privacy and access controls. Secure, Low-Latency Alerting: Building an encrypted server pipelines that delivers emergency notifications in seconds, balancing rapid response times with strict privacy and access controls. On-Device Performance Constraints: Running continuous audio and vision monitoring on a mobile device forced us to optimize heavily for battery, latency, and resource usage while maintaining real-time responsiveness. On-Device Performance Constraints: Running continuous audio and vision monitoring on a mobile device forced us to optimize heavily for battery, latency, and resource usage while maintaining real-time responsiveness. Stealth-First UX Engineering: Creating a decoy-to-secure interface that looks completely ordinary while still enabling instant secure transitions. This required careful state management so protection is seamless without exposing user intent. Stealth-First UX Engineering: Creating a decoy-to-secure interface that looks completely ordinary while still enabling instant secure transitions. This required careful state management so protection is seamless without exposing user intent.

Accomplishments we're proud of

Built a stealth-first safety app designed to protect users in real-world emergencies Achieved sub-second semantic threat detection and autonomous escalation Implemented decoy-to-secure mode switching for discreet privacy protection Enabled always-on on-device audio and vision monitoring without user input Integrated OpenAI, Supabase, and Twilio into a complete end-to-end safety platform

What we learned

Designing stealth-first UX where safety depends on seamless "decoy-to-secure" transitions Deploying real-time on-device computer vision for autonomous privacy protection and auto-locking Building low-latency continuous audio threat detection pipelines for high-stakes environments Making AI systems trustworthy by combining structured LLM reasoning with deterministic escalation logic and human-in-the-loop control Designing a reliable full-stack architecture

What's next

We can bolster the app by developing the following features: Improving AI reliability using personalized risk calibration and stronger safeguards against false positives Adding end-to-end encryption and a hardened vault for evidence storage and legal documentation Building a trusted network mode where verified contacts or advocates can receive real-time escalation updates Upgrading emergency workflows, including silent check-ins, timed safe-word triggers, and automatic location sharing

Analysis

Compare with all teams

View

Metric

Figures cover GitHub contributors during the hackathon window. A co-authored commit counts in full for each author, so per-member totals add up to more than the whole-team figures.

Technology

Found in codeClaimed only

SwiftIn code
OpenAIClaimed
SupabaseClaimed

1 of 3 appear in the indexed code. 2 claimed on Devpost could not be matched to code, which may simply mean the tool leaves no trace in the repository.

AI coding agents

No AI coding agent signals were found in this repository.

Detected from committed agent config files and commit authorship. Absence of a signal is not proof an agent was unused.

Codebase size

Source size

157 KB

Source files

Counts recognized source files only; vendored directories, binaries and lockfiles are excluded, so this is smaller than the repository on disk.