Methodology

How we ship.

Small team. Real users. Production from day one. Here's how SOYL AI builds — and what we refuse to build.

The Loop

Build → Ship → Learn

01
Log // Phase I

Phase I: Listen

Every product starts with a real person, a real problem, a real conversation. Before any code, we sit with hotel staff, with home users, with the people who'll use what we build.

Key Deliverables

  • On-site interviews and shadowing
  • Problem decomposition before solutioning
  • Hard "no" on building for personas we have not met

If we cannot describe the user in one sentence and their problem in another, we are not ready to build.

02
Log // Phase II

Phase II: Architect

We design for production from day one. Privacy-first by default. Local-first where the hardware allows. Cloud only where it earns its place.

Key Deliverables

  • Cost and latency budgets set before code
  • Privacy and data flow diagrams reviewed
  • Hardware and infra choices justified in writing

Architecture decisions get documented. Future-us deserves to know why past-us made the call.

03
Log // Phase III

Phase III: Ship

We ship to real users early. Butler AI was in a real hotel before we built half its features. The user is the spec.

Key Deliverables

  • Pilot deployments before polish
  • Production telemetry from day one
  • Bug-fixes and shipping take priority over new features

A product not in someone's hands is not a product. It is a slide deck.

04
Log // Phase IV

Phase IV: Iterate

Software rots unless it learns. Every release is informed by what last week's release taught us — from production logs, from user calls, from the team's own dogfooding.

Key Deliverables

  • Weekly review of production data
  • User calls every cycle, not just at launch
  • Roadmap reshaped by what we learn, not by what we promised

Plans are guesses. Production is the truth. We update our beliefs.

Core Ideology

The Principles We Build On

Build, then talk

We ship before we pitch. Pilots in production before pages on a website.

Iterate ruthlessly

Every release teaches us something. We update our beliefs and the product on the same day.

Local-first by default

Where it can run on your hardware, it does. Privacy is an architecture decision.

Craft the small things

A loading state, a voice latency, a mint border — the small things compound into the feel of the product.

Inside the Lab

Atmosphere

A fractional glimpse into the spaces, structures, and visual artifacts that directly shape our engineering culture.

Bengaluru HQ

A small team, a shared room, no cubicles. Where most of our shipping happens.

Grid View

Node Networks

Cross-team daily syncs.

Iteration over perfection — across every layer of the stack.

Hardware Lab

Where Jetson Orin nodes get prototyped, benched, and sometimes broken.

Architecture

Stack & practices

A small set of opinionated, production-grade tools — chosen for cost, latency, and reliability. We pick boring where we can, sharp where we have to, and we own our infrastructure end-to-end.

FastAPI + LangGraph for multi-agent orchestration
PostgreSQL + pgvector for retrieval and memory
Asterisk PBX + Twilio SIP for sub-3s voice
AMD Ryzen AI NPUs for on-property edge inference
Next.js 16 + Tailwind v4 for every product surface
Self-hosted on Hetzner — cost-efficient by design

Want to talk?

Whether you run a hotel and want to pilot Butler AI, you're a developer interested in AI Dex, or you just want to say hi — we read every message.

Get in touch →