The latest articles on DEV Community by TOLUWALASE OYEWOLE (@tolex081). https://dev.to/tolex081 https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4164980%2F75d9fcfa-74f3-4980-816a-fd321faeb5db.jpg https://dev.to/tolex081 en TOLUWALASE OYEWOLE Mon, 05 Oct 2026 22:47:13 +0000 https://dev.to/tolex081/i-built-vitarecall-a-health-ai-companion-that-remembers-without-pretending-to-be-a-doctor-45dc https://dev.to/tolex081/i-built-vitarecall-a-health-ai-companion-that-remembers-without-pretending-to-be-a-doctor-45dc <p>Health is too serious for an AI to pretend to be a doctor.</p> <p>That is why I think health-related AI should start with a more careful role: helping people prepare for care, not diagnosing them, prescribing medication, or replacing emergency services.</p> <p>I built <strong>VitaRecall</strong>, an AI care-preparation companion that helps patients:</p> <ul> <li>prepare questions before an appointment</li> <li>organise notes and care preferences</li> <li>keep track of follow-up tasks</li> <li>continue a conversation without repeating what matters to them</li> </ul> <p>VitaRecall is built with Gemini and uses <strong>Walrus Memory on mainnet</strong> for patient-controlled continuity across new chats, browsers, devices, the web app, and Telegram.</p> <h2> How memory works </h2> <p>A user can share a preference such as:</p> <blockquote> <p>“Use short, plain-language answers and remind me to write down questions for my appointment.”</p> </blockquote> <p>VitaRecall archives that context automatically by default. When the user returns in a fresh conversation, the app searches that user’s own memory namespace, retrieves relevant saved context, and uses it to make the response more helpful.</p> <p>The interface distinguishes a pending save from a confirmed Walrus receipt. Users can review storage receipts, open confirmed blob IDs in Walrus Scan, and pause future saving if they choose.</p> <p>So far, VitaRecall has generated <strong>35 unique confirmed Walrus mainnet blob IDs</strong> across its live app records.</p> <h2> Safety boundaries matter </h2> <p>VitaRecall is not a medical professional, diagnosis tool, prescription service, or emergency service. It is designed to support care preparation and continuity—not make clinical decisions.</p> <p>The current public demo uses fictional information only. A key design goal is making memory useful while giving users a clear view of what is being saved and when it is recalled.</p> <h2> What I learned </h2> <p>Persistent memory changes the usefulness of a chatbot. The challenge is not merely storing text—it is making recall relevant, visible, user-controlled, and safe enough for a sensitive context.</p> <p>I would genuinely value developer feedback:</p> <ul> <li>How would you design patient-controlled AI memory?</li> <li>What should a user be able to inspect before trusting a returning AI companion?</li> <li>How would you improve the boundary between useful support and unsafe health advice?</li> </ul> <p>Live app: <a href="https://vitarecall.vercel.app" rel="noopener noreferrer">https://vitarecall.vercel.app</a><br> Telegram: <a href="https://t.me/VitaRecall_bot" rel="noopener noreferrer">https://t.me/VitaRecall_bot</a></p> <p>Source code: <a href="https://github.com/Tolex081/vitarecall" rel="noopener noreferrer">https://github.com/Tolex081/vitarecall</a></p> <p>Demo walkthrough: <a href="https://tinyurl.com/VitaRecall" rel="noopener noreferrer">https://tinyurl.com/VitaRecall</a></p> <p>Built for Walrus Sessions 8: Chatbots That Remember.</p> ai webdev programming productivity