FamCare - Healthcare management RAG

Rag System Diagram

Problem

With the Aging population, we are seeing more and more decline in fertility rate, increase in chronic diseases, and increase in healthcare costs. More Importantly, we are seeing a decline in the quality of life, which is leading to more and more people being dependent on others for their daily needs. Now we have a situation where younger ones responsible to look after older family members. Older family members are facing Chronic diseases like diabetes, hypertension, heart diseases, etc. This requires keep track of complex drugs, schedules, and a healthy lifestyle. The need to have better diet plans, exercises, medicine schedules, keeping track of health records, track sleep patterns, and organizing family activities to avoid loneliness is becoming more and more important.

Introducing FamCare

The Main objective of famcare is to help you organise your health documents, ai chat to better understand records, create medicine scheduling using ai, keeping track of medicines, schedules, diet, exercise and create a calendar, that can be shared with family members. You can download/ add events to your own ics supported calendars like apple, google, notion.

🔗 Open Source & Demo

Source Code (Open Source): https://github.com/biomathcode/famcare
Live Demo: https://famcare-one.vercel.app

Fam Care Calendar Dashboard

Story behind building this app

i moved back my parents house, and i have become responsible for managing family health, medicines, diet and exercise plans. I wanted to build an app which can help me manage all these things in one place. I also wanted to use AI to help me with the same. AI can be a great organizing tool as well as a research tool to find information about medicines, side effects, interactions etc. Also to save medical history, records and their medicine Schedules.

Features:

Add Family Members
Add Medicines
Add Health Records/ docs/ pdf
Add Medicine Schedule,
Add Diet,
Add Exercise Goals,
Add Sleep Goals.
Chat Features.
Create a Medicine Schedule Calendar.

Tech Stack

Frontend: React, shadcn, tanstack-router, tailwind
Backend: trpc, drizzle, zod,
Database: tidb
AI: moonshot ai kimi k2
Vector DB: tidb,
embeddings : jina ai,
parser : jina ai

How the RAG System work?

This is a health management system that uses RAG to answer questions about health. We upload the file then create chunk, embedding, and this become the knowledge base of our rag. We create and save vectors in tidb of length 768.

I created a custom type for vector in drizzle


export const vector = customType<{
    data: number[];
    config: { length: number };
    configRequired: true;
    driverData: string;
}>({
    dataType(config) {
        return `VECTOR(${config.length})`;
    },
    toDriver(value: number[]) {
        return `[${value.join(",")}]`;
    },
    fromDriver(value: string) {
        try {
            return JSON.parse(value) as number[];
        } catch {
            return value
                .replace(/^\[|\]$/g, "")
                .split(",")
                .map(Number);
        }
    },
});

Here is a Diagram of the rag system powered by jina ai, moonshot ai and tibd

Rag System

As you can see we have different ways of parsing image and pdf/docs. We use the moonshot-v1-8k-vision-preview for getting the readable text from the image and we use jina reader api to get the text from pdf Documents. Then We use the segment from jina ai to create chunks and the jina-embeddings-v3 to create embeddings.

Retrieving Doc Chunks using the vector search by tidb

🔍 Search Flow

User submits a question
Generate query embedding (Jina AI)

Perform Vector Search in TiDB:

SELECT id, media_id, chunk, `order`, vec_cosine_distance(embedding, ?) AS distance
FROM mediaChunks
ORDER BY distance
LIMIT 5

Retrieved chunks passed to Kimi AI for final response generation.

Tools in AI Chat

You can use the following tools to help the user:

getMembers: Use this tool to get all members from the database
createExerciseGoalTool: Use this tool to create an exercise goal for a member
createDietTool: Use this tool to create a diet plan for a member
createMedicine: Use this tool to log a medicine entry for a member
getDrugLabel: view drug labeling info
getAdverseEvents: lookup reported side effects
getDrugRecalls: check recall or enforcement reports
findRelevantContent: Use this tool to find relevant content from the database
createBulkEventsTool: Use this tool to create events in bulk or than one event

Integration with Medical APIs

Medicine search using RxTerms API
Condition search via Clinical Tables API
Real-world adverse event data via OpenFDA

We are use open fda api for getting conditions and list of medicines. We also use the tool getAdverseEvents to get the real recorded side effects of medicines.

For getting getting the list of medicines we are using the rxterms api

async function fetchMedicines(query: string) {
    if (!query) return [];
    const res = await fetch(
        `https://clinicaltables.nlm.nih.gov/api/rxterms/v3/search?terms=${encodeURIComponent(
            query
        )}&maxList=20`
    );
    const data = await res.json();
    // data[1] = array of names
    return data[1] as string[];
}

Similarly for getting the Conditions

async function fetchConditions(query: string) {
    if (!query) return [];
    const res = await fetch(
        `https://clinicaltables.nlm.nih.gov/api/conditions/v3/search?terms=${encodeURIComponent(
            query
        )}&maxList=20`
    );
    const data = await res.json();
    return (data[3] || []).map((item: string[]) => item[0]);
}

Here is the updated and polished section to add at the end of your write-up, with proper formatting and clarity:

⚙️ Installation

git clone https://github.com/biomathcode/famcare.git
cd famcare/apps/client
pnpm install

Add a .env file inside apps/client/ based on the provided sample.env.

Here is a clear and structured documentation section for setting up the database schema on TiDB using Drizzle ORM in your project:

🗄️ Database Schema Setup with TiDB & Drizzle

FamCare uses Drizzle ORM to manage the database schema on TiDB.

Step 1 – Configure Drizzle

In your project, the database schema is defined using Drizzle's schema DSL (e.g., inside src/lib/db/schema/).

A typical schema file looks like this:

import { mysqlTable, varchar, int, timestamp, json, customType } from 'drizzle-orm/mysql-core';

export const member = mysqlTable('member', {
    id: varchar('id', { length: 36 }).primaryKey(),
    name: varchar('name', { length: 255 }).notNull(),
    dob: timestamp('dob').notNull(),
    conditions: json('conditions').notNull(),
    // ... other fields
});

Step 2 – Drizzle Kit CLI

Your package.json includes a script to manage the database schema:

"scripts": {
    "db": "drizzle-kit"
}

This uses drizzle-kit for migrations.

Step 3 – Running Migrations

Make sure your .env file contains the correct database connection details:

   DATABASE_URL=mysql://user:password@host:port/database

Generate a migration based on your schema files:

   pnpm run db generate

Apply the migration to your TiDB database:

   pnpm run db push

Step 4 – Inspect & Manage Migrations

You can inspect generated migration files in the drizzle/migrations directory.

To reset the database or apply migrations from scratch:

pnpm run db reset

Example Workflow

# Install dependencies
pnpm install

# Generate migration from schema
pnpm run db generate

# Push schema changes to TiDB
pnpm run db push

The migrations are designed to be idempotent, so you can run db push safely during development.

🚀 Deployment

You can deploy the app on Vercel, Netlify, Cloudflare, or any Node server since the app uses Tanstack Start.

Vercel

Set target: "vercel" in vite.config.ts.
Deploy with:

   npx vercel deploy --prod

Cloudflare

Set target: "cloudflare-module" in vite.config.ts.
Deploy with:

   npx wrangler deploy .output/server/index.mjs --assets .output/public

Netlify

Set target: "netlify" in vite.config.ts.
Deploy with:

   netlify deploy

This makes your submission complete, easy to follow, and ready for any judge or developer to try. Would you like me to generate a formatted PDF version with this included?

🚀 Impact

FamCare empowers families to manage complex healthcare needs with ease, improving the quality of life for older adults while easing the caregiver's burden. It brings medical records, scheduling, and health advice into one place, supported by powerful AI tools.