Skip to content
Mahesh Kumar Kanthaiya — home

Case study

Status: Live

Dinelax POS — a restaurant platform built solo with AI

Role
Founder & Lead Engineer
Timeline
2025 – present
Team
Sole engineer
Markets
India and the UAE
  • Flutter
  • Dart
  • Riverpod
  • Firebase
  • Hive
  • Node.js
  • Windows
  • Claude Code
Get it on Google Play (opens in a new tab)dinelax.in (opens in a new tab)

Why I built it

Copy link to this section: Why I built it

I run a restaurant — Ariya Hotel in Tirunelveli — so I built the POS I needed there first.

Ariya Hotel became its first live customer and test bed. Then I took Dinelax POS to other restaurants in India and the UAE.

What I built

Copy link to this section: What I built

Five connected Flutter/Firebase apps plus a Windows desktop client, all built and maintained by me.

  • POS Dashboard

    The counter: billing, tables and reports.

  • Waiter App

    Waiters take orders at the table from a phone.

  • Kitchen Display

    Orders reach the kitchen screen as they are placed.

  • QR Ordering

    Guests scan a table QR code and order from their own phone.

  • Parcel App

    Takeaway orders placed from the customer’s own phone, ready for pickup.

  • Windows desktop client

    The POS on a Windows PC at the counter.

Architecture

Copy link to this section: Architecture

Multi-tenant on Firebase, multi-region for India and the UAE, and offline-first on every device.

01 · Devices

  • Android tablets
  • Phones
  • Windows PC
  • BLE thermal printers

02 · Five apps + Windows client

  • POS Dashboard
  • Waiter App
  • Kitchen Display
  • QR Ordering
  • Parcel App
  • Windows client

03 · Firebase

  • Auth
  • Firestore (multi-tenant)
  • Cloud Functions
  • FCM
  • Crashlytics
  • Hosting

04 · On device

  • Hive offline cache
  • Printers: Bluetooth ESC/POS, 80 mm, BLE chunking.
  • Regions: AED currency, 5% VAT and TRN-compliant invoicing for the UAE.
Dinelax POS architecture: devices run the five apps and the Windows client, which sync through Firebase and keep a Hive offline cache on each device.

How AI was used

Copy link to this section: How AI was used

Claude Code as pair programmer

I built the whole Dinelax POS suite with Claude Code as my pair programmer inside Google Antigravity IDE. AI writes code fast. My job is making sure it’s right.

Spec-first docs

PRD, architecture and task docs come before any code, so the AI works against a written plan instead of guessing.

Multi-agent audit: 11 P0 issues

I pointed AI agents at my own codebase for a full production-readiness audit. It surfaced 11 P0 issues across tenant isolation, PIN hashing, tax order, timezone boundaries, offline sync and duplicate bills.

What the audit changed

I fixed the security-rules batch first, then self-audited the fixes and closed 6 further holes. The remaining audit items are being closed batch by batch.

Engineering highlights

Copy link to this section: Engineering highlights

  1. PIN hashing without locking anyone out

    Salted PBKDF2 hashing with byte-identical Dart and Node.js test vectors. The switch-over was held so staff on older Play Store builds were never locked out.

  2. A live production migration

    Tenant-scoped staff records migrated in production with zero legacy rows left.

    26
    rows written
    25
    legacy rows removed
    0
    legacy rows left

  3. Thermal printing over Bluetooth

    Bluetooth ESC/POS printing on 80 mm printers with BLE chunking. Kitchen-ticket (KOT) routing was redesigned in v1.9.6 so shop-wide and per-device auto-print run together without double tickets.

  4. Built for the UAE

    Multi-region architecture with AED currency, 5% VAT and TRN-compliant invoicing. UAE client requests — day close, shift close, fund transfer, Z-report cash reconciliation — translated into scoped product features.

Quality

Copy link to this section: Quality
static analysis
Clean
Flutter tests
244
security-rules emulator tests
54
  • Closed testing → production on Google Play, with Play App Signing.
  • Backward-compatible rollouts, so older builds keep working.
  • Crashlytics triage, with fixes shipped from daily customer reports.

Results

Copy link to this section: Results
on Google Play, in restaurants in India and the UAE
Live
UAE customers
5
inbound restaurant leads from Meta click-to-WhatsApp campaigns
150+

What’s next

Copy link to this section: What’s next
  • Closing the remaining audit items, batch by batch.
  • Dinelax Boss — a Python + Flutter AI assistant for sales, support, demos and training in Tamil, Tanglish and English, built spec-first with Claude Code. In progress.

Want this kind of engineering on your team?

Open to AI Solutions Developer / AI Engineer roles in Abu Dhabi. Available immediately.

Get in touchDownload CV