How I built a production SaaS entirely from an Android phone

Prashant — founder, Sonorix · July 2026 · Mumbai

I don't own a laptop — not as a stunt, just how it worked out, and then it became the whole way I build. Everything behind Sonorix — the Android app, the backend, the payments, the server it all runs on — was written, deployed, and is operated today from the phone in my pocket.

Not "I sometimes fix a typo from my phone." All of it. ~34,000 lines of Kotlin across 298 files and 66 Activities, a Python backend, nginx configs, database migrations, a payment integration handling real money. Written with two thumbs and a terminal.

the setup

Termux gives me a real Linux userland on Android: bash, git, vim, ssh, python, and — the part people don't believe — a full JDK and Gradle. The Android app builds on the device itself. The backend lives on a small VPS I SSH into; the phone is both the development machine and, during app builds, the build server.

surviving your own fat thumbs

Editing production code on a 6-inch screen means one mistargeted tap can wreck a file, so the pipeline assumes I will screw up. Every file edit goes through the same ritual: locate the file by content (never trust a filename), take a SHA-256 of it, back it up to a manifest, write the change, hash it again, and log timestamp / file / pre-hash / post-hash / reason to a running manifest. Every deploy script prints its own revert command before it exits. Paranoia as architecture.

the stack

App: Kotlin, classic XML Views (I rolled back a Compose migration — build times on-device made it a bad trade). Ships as a direct APK from the site with its SHA-256 published beside the download. Backend: Python behind gunicorn and nginx on Ubuntu, SQLite for the main store with Postgres alongside for the queue. AI layer: Whisper for call transcription and Llama 3.3 70B for the analysis — extracting intent, budget, objections, and follow-ups from sales calls, tuned per industry vertical. Payments: Razorpay, one-time credit packs.

honest numbers

Because the internet has enough inflated launch posts: ~1,600 signups, ~40,000 analyses run, and a first handful of paying customers as of last month. Revenue so far is beer money. The product works; the current battle is activation, not features. That's the real state of it.

what a phone-only workflow actually costs

Builds are slow. Long files are navigated by search, never by scrolling. Copy-paste buffers are tiny, so anything big moves as files, not clipboard. You develop an almost religious relationship with scripts, because every manual step is ten times more painful than automating it. And in exchange: my entire company boots in the time it takes to unlock a phone, from anywhere I'm standing.