Safu Box

Workplace Harassment Statistics: Women Worldwide

• 23% of employed individuals have experienced workplace violence or harassment.
• 44% of women surveyed experienced harassment or micro-aggressions.
• 54% of women have faced sexual harassment at work, with many not reporting incidents.

Women often struggle to react to sexual harassment in the moment due to overwhelming feelings of fear and anxiety

Studies show that while women predict they would confront a harasser, in reality, many do not take action, often opting for silence or avoidance instead.

For instance, one study found that none of the women confronted their harasser during a simulated interview scenario, despite anticipating they would do so. This disconnect highlights the significant emotional barriers women face when confronted with harassment.

Our Solution:

Safu Box is a platform designed to empower individuals with tools to record, verify, and resolve disputes or conflicts ethically, securely, and privately. We integrated AI-powered wearables, a simple mobile app, and on-chain proof. Safu Box ensures that critical moments in conversations can be documented and verified without compromising the privacy of those involved. We aim to address harassment, verbal agreements, or interpersonal conflicts, Safu Box provides a robust framework for accountability and dispute resolution.

The device passively listens to conversations to detect specific keywords, tones, or patterns indicating disputes or critical moments. Events are flagged automatically, but raw audio/transcript is never stored outside of user device, ensuring complete privacy. Users can manually manage moments they consider significant. The AI focuses on extracting relevant insights without storing unnecessary context.

Here’s a detailed breakdown of how we built it

Hardware:

We use OMI device, as it is an open-source AI wearable that can be extended easily for any audio related use cases.

Mobile Application:

The mobile app is built using React Native, chosen for its ability to create cross-platform applications with a seamless user experience on both iOS and Android. All conversation data is stored offline in a SQLite database. This approach ensures reliable and private access on device only by the owner.

The backend is built in TypeScript, running on Bun. Conversations are uploaded to the backend, where they are processed using OpenAI’s GPT models.

AI models detect: Harassment and Offensive Content: Identifying harmful language or potential violations. Sentiment and Contextual Analysis: Extracting meaningful patterns or risks to empower moderators.

Proof and Immutability: When offensive content is detected, we generate a cryptographic proof/checksum to validate the authenticity of flagged content. The proof is stored on-chain via smart contracts (detailed below), ensuring tamper-proof, immutable records of flagged conversations.

Smart contracts are written in Solidity and deployed on a blockchain. Each proof/checksum is stored on-chain to guarantee:

• Immutability: Content cannot be altered or deleted, even by malicious actors.
• Transparency: Ensures trust and verifiability for legal compliance and audit purposes.

Only non-identifiable metadata and proofs are stored on-chain to maintain user privacy while ensuring security.

Github Frontend: https://github.com/Ferossgp/the-network-state Github Backend: https://github.com/anastasiarods/eth-bangkok-server