Eradicating Domestic Violence with Award-Winning UNDERCOVER App Powered by Artificial Intelligence (AI)

In Hong Kong, hundreds of women are affected by gender-based violence and domestic abuse every day. Extensive studies have shown that there is a common thread running through these incidents—verbal insults are often uttered by perpetrators before physical assaults. Of concern is that such incidents often go unreported because there is no evidence.

Recognizing the urgency to address these problems, Home-grown non-profit organization “Women Helping Women Asia” sought our help via Ogilvy Hong Kong to create an app to detect, record, and store evidence of verbal abuse that is legally admissible in court.

AI Solution for a Mobile Violence Detection App

UNDERCOVER is the first mobile app whose AI system has been taught to recognize linguistic patterns that indicate insults and profanities in English and Cantonese.

Using artificial intelligence and an open-source machine learning framework, the app has been trained to recognize specific verbal insults and distinguish them from normal speech patterns and background noise. When an insult is heard at more than 90 decibels, the app secretly records the entire subsequent incident. Using deep learning method, the AI algorithm can determine when the incident escalates to a potentially life-threatening situation, regardless of the perpetrator’s tone of voice.

In the event of an escalating incident, the app’s built-in SMS feature facilitates emergency alerts and notifications to a pre-determined trusted contact.

Data Training

In the initial phase, the app was trained with data in Cantonese and English. With access to suitable datasets in each local language, Verticurl was able to fine-tune the AI language models for optimal performance over a two-month period. These language models were trained with massive amounts of data, enabling them to generate contextually relevant text with accuracy.

Regular QA testing throughout development allowed us to verify the functionality, usability in different scenarios, accents, and consistency across multiple mobile devices, as well as identify vulnerabilities in the app’s security measures. This included rigorous testing with different background sounds to ensure that the app worked optimally in real-life situations.

The final AI language model consists of more than 20 classifications of targeted phrases with high recognition accuracy.

Demo Film: Example of a specific phrase initiating a recording

Discreet Design, Biometrically Protected

As the name suggests, the design of the UNDERCOVER app is clean and minimalist to avoid detection. It relies on biological authentication methods such as fingerprints and facial recognition to provide victims with an extra layer of security to access the app, making it difficult for the perpetrator to access the recording—even if the device is stolen. The app has key features where it’s needed most, and an intuitive user interface, designed to prevent accidental touches from stopping the recording.

A recording library has been installed to discreetly log incidents of verbal abuse with the exact time, date and location, and the app keeps it secure by using biometric activation for every user login.

An Award-Winning App That Inspires Real Action

Mobile apps have changed the way we all live but very few have empowered victims of gender-based abuse to take control of their plight and take real action.

UNDERCOVER is an initiative of Women Helping Women Asia—a collective of organizations that provide tools and resources for victims of domestic violence, single mothers, and other women in need.

• Winner of the Graphite Pencil 2023 in the Digital Design/Health & Wellbeing category, D&AD Award
• Winner of 2 Bronze Lions 2023 at Cannes Lions International Festival of Creativity in the Design and Digital Craft category

UNDERCOVER is available for free on Google Play Store and is distributed through community groups and women’s organizations throughout Hong Kong. It’s available in Mandarin, Vietnamese, Korean, Japanese, and other languages.

Credits: Ogilvy Hong Kong & Verticurl Vietnam