Sahayak AI

Inspiration Healthcare access is incredibly fragmented, especially for underserved and middle-class families. We realized that caregivers often act as the "unofficial doctors" of their households, constantly Googling symptoms while struggling to remember how a new issue might interact with a grandparent's blood pressure medication or a child's allergy. Generic AI chatbots are dangerous because they lack this vital context. We were inspired by UN SDG 3 (Good Health) and SDG 10 (Reduced Inequalities) to create a tool that brings enterprise-grade, context-aware triage directly into the homes of those who need it most.

What it does Sahayak AI is an intelligent, family-centered health copilot. Instead of starting from scratch every time you have a health question, Sahayak acts as a unified household dashboard. Users create lightweight medical profiles for their family members—logging conditions, allergies, and daily medications.

When a concern arises, you simply select that family member and describe the symptoms. Sahayak uses IBM watsonx.ai to analyze the issue specifically against that family member's pre-existing profile. It then generates a safe, structured response card that highlights the urgency level, suggests a care route (e.g., self-care vs. emergency room), and explains the reasoning. Every interaction is automatically logged into a chronological health timeline to share with actual doctors later.

How we built it We architected the platform prioritizing speed, UI excellence, and AI safety:

Frontend: We used Next.js (App Router) and Tailwind CSS to build a highly responsive, accessible, and premium UI. Backend & Database: We utilized Next.js Server Actions and API routes connected to a Prisma ORM (with SQLite for the MVP) to handle relational data between families, profiles, and event logs. AI Engine: IBM watsonx.ai is the brain of the application. We built a dedicated integration layer (lib/watsonx.ts) that intercepts the user's query, dynamically injects the patient's medical context into the prompt, and enforces a strict JSON output schema to ensure the AI's advice remains safe, structured, and non-diagnostic. Challenges we ran into One of the biggest challenges was ensuring AI safety. LLMs are prone to hallucinating medical advice or acting with too much certainty. We had to carefully engineer the prompts sent to IBM watsonx.ai to ensure it acts as a triage navigator rather than a doctor. We spent significant time forcing the model to output structured JSON so we could reliably render visual "Urgency" badges (Emergency, Urgent, Routine) and strictly include medical disclaimers in the UI.

Accomplishments that we're proud of We are incredibly proud of the contextual RAG-style pipeline we built using IBM watsonx. If you ask a generic AI about "dizziness," it gives a generic answer. If you ask Sahayak AI about a grandfather's dizziness, the IBM engine explicitly cross-references his profile, notes that he takes Metformin, and accurately warns about potential hypoglycemia. Seeing the AI safely catch those contextual medication overlaps was a massive win for the project's viability.

What we learned We learned a massive amount about prompt engineering for safety and how to effectively structure unstructured LLM outputs for a React frontend. We also learned how powerful Next.js Server Components are for keeping API keys (like our IBM Cloud credentials) entirely hidden from the client while maintaining a blazing-fast user experience.

What's next for Sahayak AI We plan to deploy the architecture to IBM Cloud Code Engine. For feature expansion, we want to implement multi-lingual support via IBM Watson Speech-to-Text so rural users can ask questions verbally in their native languages. Finally, we aim to integrate an OCR pipeline to allow families to snap pictures of physical lab reports and have watsonx.ai summarize the results directly into their permanent timeline.

Built With

• accessing-immediate
• api
• ibm
• mern
• non-diagnostic-triage-response?telling-you-whether-it's-safe-to-rest-at-home
• often-forgetting-how-a-new-symptom-might-interact-with-an-existing-allergy-or-a-daily-medication-like-metformin."-the-solution-(what-it-is):-"that?s-why-we-built-sahayak-ai?a-family-centered-health-copilot.-instead-of-a-generic-chatbot
• or-if-you-need-to-escalate-to-an-emergency-room-immediately."-the-impact-(un-sdgs):-"by-directly-aligning-with-un-sdg-3-(good-health)-and-sdg-10-(reduced-inequalities)
• prevents-dangerous-drug-interactions
• reliable-healthcare-is-a-luxury-they-don't-have.-caregivers-are-left-googling-symptoms
• sahayak-acts-as-a-unified-household-dashboard.-it-maintains-lightweight-medical-profiles-for-everyone-in-the-family?from-a-child's-peanut-allergy-to-a-grandfather's-diabetes."-the-magic-(ibm-tech):-"when-a-concern-arises
• sahayak-ai-bridges-the-gap-between-uncertainty-and-professional-care.-it-reduces-unnecessary-clinic-visits
• the-last-thing-you-want-to-do-is-guess.-but-for-millions-of-underserved-households
• you-simply-'ask-for-this-member.'-we-use-ibm-watsonx.ai-to-analyze-the-new-symptoms-strictly-within-the-context-of-that-specific-family-member's-pre-existing-conditions.-the-ibm-engine-instantly-generates-a-safe