Hearty

AI concierge MVP for active seniors

Available at: d2jc32x9xxe9d3.cloudfront.net

Issue / Solution

The Issue: Many active seniors and adult learners start lifestyle courses (like classical instruments, golf, or cooking) with high enthusiasm but drop out within a few weeks. Standard directories or marketplaces offer hundreds of scattered listings but fail to support the end-to-end learning journey, leaving seniors feeling overwhelmed, unguided, and disconnected during their initial lesson stages.

The Solution: I built Hearty, a deployed web MVP focusing on course completion rather than simple registration. Seniors are onboarded through a conversational AI voice assistant (Evelyn) that conducts a comfortable diagnosis to extract schedules and barriers. The system seeds a structured 12-session learning workspace with active milestones, manager notifications, lesson check-ins, and direct message lines, deployed using Next.js on CloudFront and an AWS Chalice/Neon Postgres backend.

Overview

Hearty is a deployed web MVP for Korean adult learners and active seniors. The product combines a polished public landing page, an onboarding diagnosis flow, sample login, dashboard screens, schedule, records, messages, notifications, and backend-backed user data. The current deployment is live through CloudFront and backed by an AWS Chalice API.

System Architecture

The system is built static-first for ultra-low hosting costs, leveraging AWS serverless compute and serverless relational storage.

System Architecture

Product Flow

The landing page frames the product around completion rather than simple registration: AI helps a user find a suitable activity, and a dedicated manager supports the 3-month journey. The diagnosis page introduces Evelyn, the AI concierge, as a conversational onboarding agent. After signup or sample login, the app can seed a starter workspace with a plan, lessons, records, messages, and notifications so the dashboard feels like an actual service instead of an empty prototype.

Landing Page & Conversational Onboarding

Hearty's client interfaces use warm, highly readable layouts tailored for elder accessibility. They feature large font ranges and a relaxed, voice-guided conversational flow.

Hearty deployed landing page — **Eldercare Landing Page (시작은 늘 했습니다. 이번엔, 끝까지):** Focuses on course completion. Active seniors are welcomed with a clear call-to-action button (무료 진단 시작) to kick off the conversational diagnostic onboarding.

Hearty AI concierge diagnosis screen — **Conversational Onboarding Concierge (에블린과의 대화):** Driven by Evelyn, the AI agent. Users can trigger a comfortable 5-minute mic conversation (마이크를 눌러 에블린과 대화를 시작하세요). Evelyn analyzes audio files dynamically in the background to suggest a customized course plan.

What I Built

A static-export Next.js 16 frontend hosted from private S3 through CloudFront.
A serverless AWS Chalice backend deployed through API Gateway and Lambda in Seoul.
Neon Postgres schema and migrations for users, profiles, plans, lessons, activity logs, messages, notifications, OAuth tickets, and sessions.
JWT signup/login, sample-account login, Google OAuth flow, and Kakao OAuth groundwork.
Live dashboard routes for home, schedule, records, messages, notifications, and diagnosis-result save.
AI concierge screens and prompt design for Evelyn, including voice-diagnosis structure and user-profile extraction.

Architecture

Hearty is intentionally static-first. The frontend builds into plain files under frontend/out, uploads to private S3, and is served through CloudFront. Browser requests call the API Gateway endpoint, which invokes a Chalice Lambda connected to Neon Postgres. This keeps hosting simple and inexpensive while still supporting live authentication and dashboard data.

The backend API includes health checks, signup/login, sample login, OAuth start/callback, current-user lookup, dashboard data, plans, lessons, activity logs, messages, notifications, and diagnosis-result saving. Resource reads follow a tenant-isolation rule: if a logged-in user should not see a resource, the API returns 404 rather than revealing whether that resource exists.

Frontend

Next.js static export

CloudFront + private S3 delivery for landing, onboarding, auth, and dashboard pages.

Backend

Chalice Lambda

API Gateway, JWT auth, sample workspaces, live records, messages, and notifications.

Operations

The repo includes deployment scripts for provisioning the frontend bucket, uploading the static build, invalidating CloudFront, and deploying the Chalice backend. The README also records the current CDN URL, backend API URL, AWS region, and Neon database setup so the project can be handed off or resumed without rediscovering infrastructure details.

Stack

Next.js 16, React 19, TypeScript, AWS S3, CloudFront, API Gateway, AWS Lambda, AWS Chalice, Python 3.12, Neon Postgres, JWT, Zustand, TanStack Query, Google GenAI.

(go back)