Inspiration
Online misogyny is a pervasive problem severely impacting gender equality and reducing online safety (SDGs 5 & 10). Existing tools are often reactive. We were inspired to create a proactive solution that helps users recognize and prevent harmful language in real-time, fostering safer digital spaces from the source.
What it does
ShieldHER is a real-time misogyny detection simulator. As a user types, it analyzes text using an AI model, provides instant visual flagging (highlighting, alerts), displays a likelihood score with severity, and offers actionable suggestions for improvement with one-click application. It also includes integrated statistics to educate users on the impact of online misogyny.
How we built it
We built ShieldHER with a Next.js App Router frontend styled using Tailwind CSS for a clean, responsive UI. The backend is a Python/Flask API that uses a pre-trained transformer model (annahaz/xlm-roberta-base-misogyny-sexism-indomain-mix-bal) for detection. We simulated the 'real-time' feedback using debouncing on the frontend API calls to the backend.
Challenges we ran into
We built ShieldHER with a Next.js App Router frontend styled using Tailwind CSS for a clean, responsive UI. The backend is a Python/Flask API that uses a pre-trained transformer model (annahaz/xlm-roberta-base-misogyny-sexism-indomain-mix-bal) for detection. We simulated the 'real-time' feedback using debouncing on the frontend API calls to the backend.
Accomplishments that we're proud of
We are most proud of successfully building a working, real-time simulator that integrates an AI model to address a critical social issue within the hackathon timeframe. Creating a polished and intuitive UI with features like suggestions and integrated impact statistics is another significant accomplishment.
What we learned
We learned a lot about the complexities and nuances of detecting subjective language like misogyny using ML. We gained insights into effectively tuning pre-trained models for specific tasks and the importance of pragmatic solutions like debouncing for simulating real-time experiences in web applications.
What's next for ShieldHER
Our vision is to develop ShieldHER into a cross-platform Browser Extension for seamless, always-on protection wherever users type online. We also plan to expand detection to include text from URLs and Images (via OCR), continuously improve the ML model accuracy, and explore API integrations and partnerships with NGOs to scale the impact.
Log in or sign up for Devpost to join the conversation.