HerBloom

Inspiration

The idea for HerBloom started with a simple, frustrating observation: the healthcare system is exceptionally good at procedures, and poor at everything that comes after them. Women undergoing C-sections — one of the most common major surgeries performed in the United States — are routinely discharged with a pamphlet and a follow-up appointment 4–6 weeks away. Women leaving the hospital after a miscarriage are often given even less.

We are a team of four students attending a women's health hackathon, and we kept coming back to the same question: what happens to these women in that gap? The answer, backed by research, is that complications develop, anxiety spikes, grief goes unacknowledged, and questions go unanswered — often at 2am when no doctor's office is open.

That gap is not acceptable. HerBloom is our answer to it.

What it does

HerBloom is a personalized post-procedure care navigation platform built for two of the most underserved recovery experiences in women's healthcare: C-section recovery and miscarriage support.

When a user opens HerBloom, they build a support profile — entering their name, age, recovery track, how many days post-procedure they are, and any emotional context they want to share. From that input, the platform generates three personalized daily care tasks powered by AI, specific to where they are in their recovery. A woman on day 5 of C-section recovery gets different guidance than a woman on day 21. A woman who notes she is feeling overwhelmed receives tasks framed around gentleness and permission, not productivity.

As she completes each task, a flower grows. A seed becomes a stem, petals open, and eventually the flower blooms in full. This is not decoration — it is the core feedback loop of the experience, a small daily signal that says: you showed up today, and that matters. For miscarriage users, the garden never dies if they miss a day. It simply waits, because healing is not linear and the platform refuses to penalize someone for having a hard day.

Below the daily care path sits a 24/7 AI chat companion. Users can ask anything — is it normal to still feel this tired?, when can I drive again?, I keep crying and I don't know why — and receive warm, evidence-based responses grounded in post-procedure recovery guidance. The companion knows what track the user is on, how far along they are, and its own limits: it always defers to care providers for clinical questions and immediately escalates emergency symptoms to urgent care.

The entire experience runs in the browser, deploys instantly via Vercel, and requires no app download, no account creation friction, and no complicated onboarding. A woman who just got home from the hospital can be using it within two minutes.

How we built it

HerBloom is a single-file HTML/CSS/JavaScript application deployed on Vercel, with AI capabilities powered by the Featherless AI API and the Meta Llama 3.3 70B Instruct model. The full stack is intentionally lean:

$$ \text{User} \xrightarrow{\text{HTTPS}} \text{Vercel (Frontend)} \xrightarrow{\text{Serverless Function}} \text{Featherless API} \xrightarrow{\text{LLM}} \text{Response} $$

Challenges we ran into

Building for women in acute physical and emotional recovery meant every design decision carried real stakes. The miscarriage track in particular required us to unlearn a lot of instincts baked into typical product thinking — no streaks, no completion percentages, no "you did it!" celebrations. Framing grief-adjacent tasks as permission rather than action required multiple rewrites of our AI system prompt. Getting the tone right was harder than getting the code right.

What we learned

This project pushed every one of us beyond our comfort zones.

On the technical side, we learned how to build and deploy a full AI-powered web application from scratch. We learned how to route API calls through serverless functions so sensitive credentials never touch the browser. We learned how to prompt an LLM to produce structured JSON output reliably, and how to build graceful fallbacks when it doesn't. We learned Git-based collaboration under pressure.

On the human side, we learned how much design choices matter when building for vulnerable users. The difference between a flower that punishes you for missing a day and one that waits for you is small in code and enormous in impact. We learned that the words an AI uses to respond to grief matter as much as the information it provides. We learned that "this is not medical advice" is not just a legal disclaimer — it is a design principle.

We also learned to cite our statistics before presenting them publicly