S.A.R.A.H. - Smart Assistant for Recovery and Health

Inspiration

From the beginning, we wanted to build something that would matter beyond a hackathon: a tool that genuinely helps people navigate healthcare and technology in their everyday lives. We thought about the people closest to us who struggle with those systems, and our grandmothers immediately came to mind. Instead of guessing, we called them and asked what they actually wanted: help with medications, simple explanations of medical information, support scheduling appointments without battling confusing portals or long calls, and an experience that respects their limitations and fears. S.A.R.A.H. (Smart Assistant for Recovery and Health) was born directly from those conversations and is named after Ardasher’s grandmother Sarah, who lives with epilepsy and played a real role in shaping what the assistant should feel like and focus on.

What it does

S.A.R.A.H. is an accessible, health focused desktop companion that helps users handle everyday health related tasks in a way that feels simple, human, and supportive rather than technical or intimidating. It can answer medical questions in clear language while encouraging users to seek professional care when needed, manage medications through reminders and structured routines, and assist with booking doctor’s appointments by connecting to services like Daylight so the user does not have to fight through complicated interfaces. It adapts to different accessibility needs, including visual impairments and epilepsy friendly interactions, with thoughtful choices around visuals, communication style, and flow, and it presents itself through a 3D character based on Grandma Sarah, which grounds the experience in a real person and a real story instead of an abstract chatbot.

How we built it

We built S.A.R.A.H. primarily in Python, using Gemini as the core AI engine and surrounding it with a set of tools that allow the assistant to interpret user intent and interact with the system on their behalf. This tooling layer is responsible for orchestrating tasks like opening the right applications, interfacing with scheduling flows, setting reminders, and reading enough context from the environment to respond intelligently without overwhelming the user. Architecturally, we treated each capability as a focused function, in spirit like composing f(g(x)), so that the system remains modular, extendable, and easier to secure. The 3D assistant interface was created from a scan of Grandma Sarah and integrated into the client, so that interaction feels like speaking with a calm, familiar guide whose presence continually reminds us of the user we are optimizing for.

Challenges we ran into

One of the biggest challenges was giving the assistant meaningful awareness of what is happening on the computer so it can help with real tasks, while still keeping behavior predictable, private, and easy to understand for non technical users. At the same time, we had to accommodate a wide range of accessibility needs inside a single product, including designing an interface and interaction style that works for people with low vision, potential cognitive overload, and epilepsy, which meant making careful choices about motion, contrast, timing, and feedback. Finally, we faced the usual hackathon constraints of time and integration complexity, needing to make fast decisions about which features to implement without sacrificing stability or the core value of the experience.

Accomplishments that we're proud of

We are proud that S.A.R.A.H. became a working assistant instead of staying a slide deck idea, and that we shipped something that can actually help users manage real tasks within the hackathon timeframe. We are especially proud of how closely the project is tied to the people who inspired it: the assistant’s identity, features, and 3D presence grew from conversations with our grandmothers and from Sarah herself, whose lived experience with epilepsy directly informed what should be included and what should be avoided. That connection kept us focused on practical impact, empathy, and dignity, and it resulted in an assistant that already feels more like a thoughtful companion than a generic widget.

What we learned

We learned that talking to real users early is not optional if you want to build something meaningful, because many of the features we assumed would be impressive were less important than simple, reliable help with everyday pain points. We saw how designing for accessibility from the start improves the product for everyone, and how complex it can be to safely let an AI act on a user’s system, which requires clear boundaries, explanations, and fallbacks. We also learned that emotional design matters: trust, warmth, and familiarity significantly change how people perceive both the risks and the value of an AI assistant, and those qualities need to be treated as core features, not decoration.