KUTUMB — A Private, AI-Powered Shared Family Memory & Coordination Hub

Nishad Bankar — Sun, 04 Oct 2026 19:23:53 +0000

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend

🏠 What I Built

Family communication is often scattered across endless WhatsApp group chats, buried calendar invites, forgotten voice notes, and hasty phone calls. When you live in a different city from your parents, keeping everyone aligned becomes even harder. My sister and I live in another city, while our parents stay back in our hometown.

I built KUTUMB to bring our family closer together—giving us a private, quiet space that remembers everything we keep forgetting.

KUTUMB is a web-first, private PWA designed as a shared family memory layer and coordination hub. It helps us:

🌉 Bridge the Distance: Gives my parents a glanceable window into our daily lives, while helping them manage local household tasks—from electrician visits to doctor appointments and grocery lists.
✈ Plan Vacations & Get-Togethers: Keeps trip plans, decisions, itineraries, and schedules consolidated in one shared, chronological timeline.
📸 Preserve Family Memories: Creates a searchable home for precious photos, notes, and milestones so special family moments don't get lost in group chat history.
🎙️ Listen to Spoken Family Digests: Generates a warm, hands-free daily bulletin read aloud using Web Speech TTS—perfect for my parents to catch up on updates during their morning tea.

🎬 Demo & Try It Live

Live Application: https://kutumb-k66g.onrender.com

⚙️ Code Repository

cyc1otr0n / kutumb-1.0

Your family's private health dashboard and shared memory layer.

🏡 Kutumb (कुटुंब)

Your Family's Shared Memory and Coordination Layer

🔗 Live Website: https://kutumb-k66g.onrender.com

📖 Overview

Family information is frequently fragmented across WhatsApp chats, sticky notes, missed calls, calendar apps, and human memory:

"When was the electrician supposed to visit?"
"Who is picking up the cake on Sunday?"
"Did Dad mention his flight time?"
"What did we decide about the weekend trip?"

Kutumb is an AI-powered shared family memory and coordination web application. It acts as a central hub where family members can share events, tasks, reminders, and preferences using natural voice or conversational text.

Powered by Google's Gemma and Gemini models through Mastra agent orchestration, Kutumb automatically parses complex, multi-intent requests, grounds factual answers strictly in stored family data, schedules durable reminders via Temporal, and delivers spoken daily audio digests via ElevenLabs and Web Speech.

✨ Key Features

🗣️ Spoken Voice Updates & Multi-Intent Parsing
- Speak naturally…

View on GitHub

🛠️ How I Built It

KUTUMB is built around open-source AI and deterministic agent orchestration rather than an unpredictable, "black-box" consumer chatbot.

Tech Stack Highlights

Gemma (Open-Weight Model): Functions as our primary reasoning engine for intent parsing, natural language extraction, and digest generation.
Mastra Framework: Orchestrates Gemma with schema-validated tools (getTodaySchedule(), getFamilyMemory(), createTask(), createFamilyEvent()).
Temporal Cloud: Powers durable background workflows for family reminders. Even if our Render web worker restarts, Temporal guarantees background reminders fire at the exact target time.
MongoDB Atlas: Stores family workspace records, event logs, task lists, and photo/document memories.
Render: Hosts our Next.js PWA web service and dedicated background worker processes 24/7.
Sentry: Delivers real-time observability, telemetry, and end-to-end agent tracing.

🌐 Why Does Open Innovation Matter?

Pasting sensitive family details—medical appointments, home gate codes, travel itineraries, and schedules—into closed consumer chatbots presents serious privacy risks, data retention issues, and potential inclusion in public training sets.

Using Gemma (Open Weights) orchestrated via Mastra transforms family coordination into a private, grounded operating system:

🔒 Strict Data Privacy: All family records remain safely inside our own MongoDB Atlas instance. Gemma only receives the specific context snippet required for a query at runtime—never storing or training on our personal lives.
🎯 Deterministic Tool Execution: Gemma acts strictly as an intent extractor calling validated Mastra tools rather than holding arbitrary database access.
✋ Confirm-Before-Write Safety: When an update contains multiple details (e.g., "Electrician tomorrow at 11 and clear the balcony"), KUTUMB generates a preview for user review before committing changes.
🧠 Zero-Hallucination Grounding: Gemma is constrained strictly by stored family records. If a detail doesn't exist, the agent answers honestly rather than fabricating dates or events.

🏆 Prize Categories Entered

Gemma: Core reasoning engine for intent classification, tool invocation, and warm family digest summaries.
Render: Production deployment for both our web client and background worker services.
MongoDB Atlas: Application database storing events, tasks, timeline entries, and image memories.
Mastra: TypeScript agent orchestration framework handling tool schema validation and context routing.
Temporal: Durable workflow orchestration for fail-safe background reminders.
Sentry: Comprehensive agent telemetry, model latency monitoring, and error tracking.

Made with ❤️ for Mom, Dad, my sister, and families everywhere.

DEV Community: Nishad Bankar

[Boost]