HerSpace is a platform empowering women through smart, AI-driven solutions. In real life, women face unique challenges—from workplace barriers and educational inequities to the demands of parenting and even domestic violence. Although many resources exist online, few truly understand a woman's unique background or offer free, personalized support.
Our platform proactively intervenes at critical moments by guiding users through personalized steps: from expressing emotions and exploring thoughts, to identifying personal strengths and mapping out actionable goals. Built with Streamlit and integrated with advanced Google products, HerSpace connects users to crisis support, therapy locators, and more—ensuring they feel seen, understood, and empowered.
We are thrilled to be part of the She Builds AI Hackathon, hosted by Google.
Check out the live project at her-space.streamlit.app (API key for testing: admin2024).
We developed this dynamic app using Streamlit, an open-source Python framework, to ensure a seamless and interactive user experience. The front-end features were implemented with HTML5 and CSS to deliver a clean and user-friendly interface.
Google Gemini: Used as the core response generator to deliver personalized insights.
Google MusicFx: Created relaxing soundtracks for a calming user experience.
Google ImageFx: Generated all static images, enhancing the visual appeal of the platform.
YouTube Data API: Provided recommended YouTube videos as additional resources.
Google Maps API: Enabled the therapy location finder, allowing users to locate mental health professionals nearby.
Choice of Language Model: Initially, we aimed to build a Retrieval-Augmented Generation (RAG) system using Google Gemini and LangChain to leverage domain-specific knowledge without the need to retrain large models. We carefully selected web content and compiled a 40-page PDF document with various solutions to issues women face. The RAG system produced highly accurate answers when the question matched content within the PDF. However, if the question varied even slightly—for example, if the user’s context was more detailed—the system struggled to respond effectively. We experimented with adjusting model parameters, such as temperature (where a temperature of 0 leads to more predictable responses and a temperature of 1 introduces greater randomness and creativity), but these adjustments did not improve the response quality. Ultimately, using Google Gemini alone provided more consistent answers. As a result, we focused on refining the prompts sent to Gemini rather than building a RAG system from scratch.
Our first design was straightforward: users would enter their challenges, and the AI would respond. But we quickly realized that this approach led to vague, overly broad answers—too general to be truly helpful. The problem wasn’t just with the responses; users’ inputs were often too vague as well. To make the experience more meaningful, we restructured the flow, adding questions that prompt users to explore their emotions, thoughts, and strengths, as well as their support networks and actionable steps they could consider. This more guided approach helps the AI give responses that feel relevant and personalized.
We also understood that some users might share deeply unpleasant experiences that could bring up painful memories. To address this, we added a “Need Supportive Response from HerSpace Agent Now” button at every step, letting users reach out for emotional support right when they need it most.
Terms of Use Page: We created a Terms of Use page to help users understand the limitations of AI-generated solutions and clarify their responsibilities in using this guidance.
Supplemental Support Tabs: In addition to AI-generated solutions, we added a Recommended YouTube Videos tab, a Crisis Support Resource page, and a beautifully designed Therapy Location Finder page (which helps users locate the nearest therapy services). These features encourage users to seek professional help when needed, providing a well-rounded support system.
Thoughtfully Designed User Interaction Flow: We invested significant effort into designing an intuitive, visually appealing interface that supports users at every step. From timely encouragement to empathetic, positive framing questions, we aimed to soothe any emotional fluctuations users might experience during interactions, guiding them to reflect on their strengths, resources, and actionable steps.
Beautifully Designed Personalized Solutions Page: The final personalized solutions page is both aesthetically pleasing and thoughtfully structured, offering users a visually rewarding and impactful conclusion to their journey.
As engineers, we used to focus primarily on website implementation and advanced models. However, in this project, we realized that what’s even more important is designing the user interaction flow and truly standing in the shoes of users to analyze their experience throughout the interaction.
Collect more data and continue exploring the implementation of a RAG model.
Gather user feedback to improve the website.