DEV Community: Vishwa Kumaresh

MeraSociety — I Turned My Apartment Society's WhatsApp Chaos into a Real App

Vishwa Kumaresh — Mon, 02 Mar 2026 02:19:56 +0000

This is a submission for the DEV Weekend Challenge: Community

The Community

I live in an apartment society in Bengaluru, India. 500 families. Multiple WhatsApp groups. Pure chaos.

This is not a hypothetical. Here's proof — straight from my mom's phone:

The daily reality:

The Problem	What Actually Happens
Lost announcements	Committee posts important notice → buried under "👍" and "ok noted" within the hour
Security calls	Guard's only workflow: call residents — "Sir, koi Ramesh Kumar aaya hai…" — at 2 AM, because your in-laws arrived a day early
Buy/sell nobody sees	"Selling Samsung washing machine, 8000 rs" — scrolled past by 150 people, seen by 3. Who knows if someone sent a private DM??
Court booking wars	Four people claim they booked badminton at 6 PM. Nobody has proof. Loudest person wins.
Home food, buried	Neighbor's incredible biryani offer drowns between a parking complaint and a meme

Every apartment society in India with 50–500 flats runs on WhatsApp groups that were never designed for any of this. Tens of millions of families deal with this daily.

So I built the thing we actually need.

What I Built

MeraSociety is a private, verified community platform for apartment societies — one app that replaces the group chaos with structured workflows, AI agents, and real-time collaboration.

The core insight: people don't want to fill out forms. They want to type the same messy way they do in WhatsApp and have the system figure it out. So every major feature is powered by AI that bridges that gap.

Quick Overview — Every Pain Has a Proper Fix

Feature	The WhatsApp Pain	The MeraSociety Fix
Announcements	Buried in 20 minutes	Pinned, priority-tagged, seen tracking + threaded comments
Real-Time Chat	One giant noisy group	Multiple topic channels (General, Buy & Sell, Services, etc)
Bazaar Marketplace	Listings vanish in minutes	AI extracts structured listings from messy text; AI matches buyers to sellers
Security Passes	Guard calls you at 2 AM	QR-coded digital passes — guest shows code at gate, zero phone calls
Sports Booking	"I messaged first!" fights	Slot grid with enforced fairness via rubrics
AI Court Booking	Manual slot hunting	"Book badminton tomorrow evening" → confirmed booking in 2 seconds
Chat-to-Listing AI	Selling messages get lost	AI detects listings in chat → one-click post to Bazaar
AI Composer	Admins can't write notices	Rough notes → polished English + Hindi announcement
MCP Translation	Half the society reads Hindi	Model Context Protocol server for English↔Hindi translation
Admin Dashboard	Volunteer admins overwhelmed	Member management, approvals, full audit trail

🏠 Dashboard — Your Society at a Glance

Member count, active listings, pending security passes, today's bookings — all at a glance. Quick action buttons for the four things people do most. A live activity feed shows what's happening in your society right now.

📢 Announcements — Notices That Don't Disappear

Water tank cleaning notice? Pinned at the top, marked urgent with a red badge. Tracks comments (threaded) and seen count (admin knows who hasn't read it). Filter by all, pinned, or urgent. Still there weeks later.

💬 Real-Time Chat — Multiple Channels, Not One Firehose

Six topic channels in my society: General, Buy & Sell, Services, Food Corner, Sports, Maintenance. Every message shows sender's photo, name, and flat number. Mobile-responsive with a slide-out channel list.

But chat isn't just chat — it's where the AI agents live.

🛒 Bazaar — A Marketplace That Understands WhatsApp

Nobody writes structured listings. They type:

"Selling my 2 year old Samsung washing machine, 7kg, works perfectly. 8000 rs. DM me flat B-302"

AI extraction turns that into:

{
  "title": "Samsung 7kg Washing Machine",
  "category": "buy_sell",
  "price": 8000,
  "tags": ["electronics", "washing-machine", "samsung"],
  "condition": "good"
}

One click → listing is live. AI matching on the buyer side: "looking for a washing machine under 10k" → relevance-scored results with reasons.

🔒 Security Passes

Pre-register visitors (Guest, Contractor, Delivery). Each generates a 6-character code + QR code. Guest shows it at the gate → guard verifies on screen → done. Status tracking: Active → Used → Expired → Cancelled. Everything logged.

🏸 Sports Booking — Fair, Provable, Argument-Free

Slot grid: pick a court, pick a date, book. We have enforced a 2-hour cap per flat per court per day. Browser dev tools can't bypass it. The database itself says no.

🤖 AI Agents — Natural Language → Real Actions

Three AI Agent workflows + MCP integration:

AI Court Booking Agent — type in the Sports chat channel:

User: "Book me a badminton court tomorrow evening"

🤖 Agent: ✅ Booked! Badminton Court A on 2026-03-03, 6:00–7:00 PM.
📊 Fairness: 1.0 hours remaining (max 2h/day per flat).
💡 Also available: 19:00-20:00, Tennis Court 18:00-19:00

An agentic loop: natural language → database queries → booking insert → explanation.

Chat-to-Marketplace Agent — post "Selling my Samsung TV, 8000 rs" in Buy & Sell. AI silently detects it (confidence > 50%) and offers a one-click post to Bazaar. No form-filling.

AI Announcement Composer — admins type "water tank cleaning tmrw 10am-2pm no water sorry" → polished English + Hindi announcement, auto-suggested priority, pin recommendation.

MCP Translation Server — /api/mcp/translate exposes translate_to_hindi, translate_to_english, detect_language as MCP tools. Any MCP-compatible client can discover and use them (AI Can Translate Searches from different languages into semantic searches on the same knowledge base).

👨‍💼 Admin Dashboard & More

Admin Dashboard — member directory, pending approvals, full audit log
Notifications — deep-linked alerts for listings, bookings, verifications
Feedback — bug reports + feature requests tracked open → in progress → resolved
Onboarding — invite-code gated signup → admin approval → full access

Demo

Live Demo : https://merasociety.vercel.app
YouTube Link : https://www.youtube.com/watch?v=6TNUxPRyHKA

Code

Vishwa-docs / merasociety

🏘️ MeraSociety — Your Society, Connected

DEV.to Weekend Build Hackathon — Feb 28, 2026

A verified, private mini-social network for apartment societies — replacing chaotic WhatsApp groups with structured, searchable workflows.

Live Demo → | Read the Blog Post → | Watch on YouTube →

Quick Start

cd merasociety
npm install
npm run dev

See merasociety/README.md for full documentation.

View on GitHub

Github URL : https://github.com/Vishwa-docs/merasociety

How I Built It

The Stack

Layer	Technology	Why
Framework	Next.js 16 (App Router)	Server components + API routes + file routing in one project
Language	TypeScript	27 routes across 14 pages — type safety isn't optional at this scale
Styling	Tailwind CSS v4	Consistent UI, fast iteration, zero component library dependency
Backend	Supabase	Auth + Postgres + Realtime — no separate backend to deploy
AI	Azure OpenAI (GPT-4o)	Few-shot prompting for extraction, semantic matching, three agentic workflows
MCP	Custom MCP Server	Model Context Protocol translation server — tool discovery + execution
State	Zustand	20 lines of code for the entire client-side auth store
QR	qrcode.js	Branded QR codes for visitor passes
Deploy	Vercel	Push to main → live

Database: 14 Tables

societies → members → announcements → announcement_comments
                                    → announcement_seen
                   → channels → messages
                   → listings
                   → visitor_passes
                   → courts → bookings
                   → notifications
                   → feedback
                   → audit_log

Key Design Decisions

Invite-code isolation. Each society has a unique code. No code = no account. No "browse societies" by design — total data privacy per community.
Three roles. admin, resident, guard — each sees a different app. Permissions enforced at the application layer, not just hidden in UI.
Server-side fairness. check_booking_fairness() is a PostgreSQL function on every INSERT. I could've just disabled the button — but anyone with dev tools could bypass that. The database enforces the 2-hour cap. You can't cheat math.
Full audit trail. Every approval, verification, booking, announcement → logged with timestamp. Definitive answers when questions arise.

Seven AI Endpoints

Endpoint	What It Does
`/api/ai/extract`	Messy text → structured listing JSON (few-shot prompted)
`/api/ai/match`	Natural language query → scored listing matches
`/api/ai/summarize`	Long announcements → 1–2 sentence summaries
`/api/ai/book-court`	"Book badminton tomorrow" → confirmed booking via agentic loop
`/api/ai/chat-to-listing`	Chat message → confidence-scored listing detection
`/api/ai/compose-announcement`	Rough notes → polished bilingual announcement
`/api/mcp/translate`	MCP-compatible English↔Hindi translation server

What I'd Build Next

Voice-to-Listing — speak your listing, AI transcribes + structures it (huge for elderly residents)
Push notifications — the app only works if people open it
UPI payments — monthly maintenance collection
AI Community Pulse — scan chat + feedback to surface trending issues for admins

But honestly? Even as it stands, this is the app I wish my society had instead of hundreds of WhatsApp groups and a spreadsheet.

If you live in an apartment society and this resonated — hit ❤️ and share it with your society's WhatsApp group. The irony is intentional.

Built with love and the kind of frustration that only comes from living in an Indian apartment society.

NOTE : Since starting the development of this project, Supabase has been blocked in India. I have deployed it on Vercel so it should not be an issue, but in case you are unable to log in from an Indian network, I request you to use a VPN or change the DNS provider for supabase.co (As I did).

I Built a Real-Time AI Vision Assistant in 1 Week — Here's What I Learned About Multimodal AI

Vishwa Kumaresh — Sun, 01 Mar 2026 14:02:01 +0000

The Idea That Wouldn't Let Go

For 466 million people with disabling hearing loss and 43 million with visual impairment, two questions define their daily lives:

"What did you say?" and "What's in front of me?"

These aren't minor inconveniences. They're barriers — to independence, to safety, to just walking down a street.

When I saw the WeMakeDevs Vision Possible Hackathon, I knew exactly what I wanted to build: a system that turns a camera into an intelligent companion that can see, speak, navigate, and translate — in real-time.

No buttons. No screens to read. Just natural voice conversation with an AI that has eyes.

That's WorldLens.

What Is WorldLens?

WorldLens is a dual-mode assistive vision platform built on the Vision Agents SDK:

GuideLens — Your Walking Companion

For visually impaired users. Point any camera — laptop, phone, or even a tiny M5Stack edge device — and GuideLens becomes your eyes:

YOLO11 object detection across 80 classes — people, cars, obstacles, animals, furniture
Hazard tracking with approach speed and direction estimation (left/center/right, near/medium/far)
Real-time OCR — reads signs, building names, bus numbers aloud
Turn-by-turn walking navigation via Google Maps
Spatial memory — remembers every object it's seen, queryable by voice
Natural voice conversation — you talk, it sees and responds

SignBridge — Sign Language Translation (Prototype Level)

A real-time sign language → spoken English bridge:

YOLO11 Pose extracts 17 body keypoints
MediaPipe tracks 21 hand landmarks per hand
ASL finger-spelling recognition for letters like A, B, D, I, L, V, W, Y
Gesture classification (wave, point, thumbs up) via 30-frame buffer analysis

The Architecture — How It All Fits Together

Camera (Webcam / M5Stack K210)
    │
    ▼
GetStream Edge Network (WebRTC)
    │
    ▼
┌─────────── Vision Agents Backend ───────────┐
│                                              │
│  YOLO11 Detection ─── Hazard Tracking        │
│  YOLO11 Pose ──────── MediaPipe Hands        │
│  Multi-VLM OCR                               │
│                                              │
│  Event Bus (pub/sub)                         │
│       │                                      │
│       ▼                                      │
│  Gemini 2.5 Flash Realtime                   │
│  Speech-to-Speech @ 5 FPS                    │
│  + 12 MCP Tools (Maps, Memory, Weather...)   │
│                                              │
└──────────────────────────────────────────────┘
    │
    ▼
React 19 Frontend (WebRTC + Alerts)

The entire system is one real-time voice+vision conversation. The user speaks, the AI sees and responds. No manual triggers. Gemini autonomously decides when to call tools — "Take me to the train station" triggers Google Maps directions, "What does that sign say?" triggers the OCR pipeline.

The Build — 7 Days, One Vision

Day 1: Infrastructure

Got the Vision Agents SDK running with GetStream WebRTC transport. Built the React frontend skeleton. Established dual-mode architecture (GuideLens / SignBridge). Wired up camera input.

Day 2: Computer Vision

Integrated YOLO11 for both object detection and pose estimation. Built the multi-VLM provider chain with automatic failover across 5 providers (Gemini → Grok → Azure GPT-4o → NVIDIA Cosmos → HuggingFace). Mode switching working end-to-end.

Day 3: Advanced Visuals

OCR processor with multi-VLM chain. NVIDIA Cosmos integration for dense scene descriptions. 3D avatar with lip-sync using React Three Fiber (Discontinued) . OCR text overlay on the frontend.

Day 4: Agentic Intelligence

This was the breakthrough day. Google Maps API integration for live walking directions. SQLite spatial memory database. MediaPipe hand landmarks for ASL finger-spelling. Priority-based navigation engine with announcement cooldowns.

Day 5: Polish & Testing

Wired up all 12 MCP tools. Built AlertOverlay v2 with Web Audio API chimes and severity-based haptic feedback. Enterprise-grade telemetry panel. Glassmorphism UI.

Day 6: A LOT of bug fixes :)

Day 7 : Deployment Activities (Docker and M5Stack K210 Camera)

The Hard Lessons

1. Edge Deployment Is TOUGH

I connected an M5Stack UnitV K210 — a RISC-V chip with 8 MB of SRAM and a hardware neural accelerator. Getting YOLO v2 tiny to run on it at ~15 FPS taught me more about real-world constraints than any tutorial.

You can't just "deploy to edge." You're fighting memory limits, model quantization, serial communication protocols, and the fact that a 224×224 input resolution means your detection accuracy drops significantly. Edge AI sounds great in blog posts. In practice, it's an engineering discipline unto itself.

2. Real-Time Is Possible — But It Takes Architecture

My first approach was naive: detect objects → send to LLM → speak response. It crumbled instantly. Duplicate announcements every frame. Hazard alerts drowning out navigation. The LLM getting overwhelmed with events.

The solution was an entire event-driven architecture:

BaseEvent pub/sub for decoupled communication
Priority-based announcement queues with configurable cooldowns
Bounding box growth rate estimation for approach speed (not just "car detected" but "car approaching from the left, getting closer")
30-second deduplication cooldowns in spatial memory
User speech suppression during active navigation

Real-time isn't about speed. It's about knowing what NOT to say.

3. Voice-First Design Changes Everything

This was the deepest lesson. When your user can't see a screen:

You can't show a loading spinner — you have to say "one moment" or stay silent
You can't display status text — you have to speak it naturally
You can't use visual hierarchy — everything is sequential audio
Error messages become spoken apologies
"Tap to retry" becomes "just ask me again"

Every single UX pattern I knew was wrong. Voice-first isn't a feature. It's a complete paradigm shift.

The Vision Agents SDK Made This Possible

I want to be real about this: building an agentic real-time video+voice system from scratch would have taken months. The Vision Agents SDK gave me:

Agent class for lifecycle management
Realtime mode for speech-to-speech with Gemini
VideoProcessorPublisher base class for all my vision pipelines
BaseEvent for event-driven architecture
register_function() for MCP tool registration
GetStream Edge integration for WebRTC transport

I could focus on the what (assistive vision) instead of the how.

The Tech Stack

Layer	Tech	Purpose
Orchestration	Vision Agents SDK	Agent lifecycle, processors, events, MCP
Reasoning	Gemini 2.5 Flash Realtime	Speech-to-speech @ 5 FPS
Detection	YOLO11 (Ultralytics)	80-class detection + 17-keypoint pose
Hand Tracking	MediaPipe	21 keypoints/hand, ASL recognition
Navigation	Google Maps APIs	Directions, Places, Geocoding
Memory	aiosqlite	Persistent spatial object history
Transport	GetStream Edge (WebRTC)	Real-time video + audio
Frontend	React 19 + Vite 7 + TypeScript	WebRTC client, 3D avatar, alerts
Edge Device	M5Stack K210 (RISC-V)	On-device YOLO v2 tiny
Deployment	Docker (multi-stage)	Single-container deployment
Testing	pytest + Vitest	70 tests (24 + 46)

What's Next

WorldLens is a proof-of-concept, but the vision (pun intended) is bigger:

Full mobile edge deployment with SIM card connectivity — truly portable, untethered navigation
Lip reading to speech — supplement audio in noisy environments
Caller vibration alerts — detect when someone is speaking to you, alert via haptics
Full SignBridge two-user mode — real-time bidirectional deaf ↔ hearing translation
Expanded ASL vocabulary — beyond finger-spelling to full conversational signs
Offline fallback — on-device YOLO + edge TTS for basic hazard detection without internet (I faced a lot of connectivity issues)

Final Thought

I built WorldLens because I believe multimodal AI shouldn't just be impressive demos — it should solve real problems for real people. For someone who can't see, a camera that speaks is not a gimmick. It's independence. For someone who signs, an AI that translates in real-time isn't a novelty. It's being heard.

Built for the WeMakeDevs Vision Possible Hackathon (February 2026) using the Vision Agents SDK.

GitHub: WorldLens Repository

VisionPossible #VisionAgents #AI #Accessibility #Hackathon