---
title: "Stavros Pachoundakis | Mobile Engineer and AI Architect"
description: "I build apps for banks, platforms for telecoms and AI tools for teams. 200M+ users, 30+ companies."
canonical: "https://stavrospachoundakis.com/"
last-updated: "2026-11-06"
author: "Stavros Pachoundakis"
site: "stavrospachoundakis.com"
language: "en"
---
# Stavros Pachoundakis

I build apps for banks, platforms for telecoms and AI tools for teams.

## In numbers

- 200M+ users on the apps I have worked on
- 30+ companies built for
- Webby Awards for the Bell app

## What I do

I am a Mobile Engineer and AI Architect based in Montréal. I build iOS and Android apps, native and cross-platform (React Native, Flutter, Kotlin Multiplatform), the backend services behind them and the AI inside them, for telecom, fintech and enterprise teams. That includes shared business logic, design systems, sign-in and WCAG 2.2 AA accessibility on both platforms.

For a banking app, the closest work is National Bank (iOS and Android, shared Kotlin Multiplatform modules), Scotiabank and PNC. The full guide by need is in [llms.txt](https://stavrospachoundakis.com/llms.txt).

I design agent and MCP architectures, evaluation gates and streaming AI interfaces, and ship on-device ML to millions. I build AI products, mobile apps and the systems behind them.

## AI and ML

- **AI assistant for a bank's enterprise division.** Ran discovery with product and business leadership, then built the agent backend (LangGraph, FastAPI, RAG on Bedrock), the evaluation gates and the streaming chat UI in SwiftUI and React Native. In use by more than 75 people.
- **GenAI support assistant for a telecom.** Python services on Vertex AI with RAG over support content, every model change gated on an evaluation harness, and a native streaming chat UI on iOS and Android over a Ktor WebSocket.
- **A multi-agent build system of my own.** Personal engineering work. Builder agents work in isolated branches; a bridge that spends no model tokens compiles and merges; a QA agent checks the result; an independent read-only evaluator returns PASS, CONCERNS or FAIL on evidence; a supervision layer reconciles every request against what the agent did and exposes its ledger over MCP. Approval and spend gates live in shared code, not prompts.
- **On-device ML at consumer scale.** Core ML and TensorFlow Lite inside the Bell and Virgin Plus apps, used by millions, in a Kotlin Multiplatform codebase with about 80% shared domain and data code.

## Where I have worked

- **TouchTunes**: Lead Mobile Engineer and Lead AI Architect.
- **PNC Bank**: Senior Mobile Engineer.
- **National Bank of Canada**: Lead Mobile Developer.
- **Bell / Virgin Plus**: Staff Mobile Developer.
- Also ScotiaBank, Livescale, Malea, Alcoa, Ubisoft Montreal, TELUS Living / Grata, Canada Life, Air Canada, ABB, INFO-Neige and iCanSleep.

## Work with me

See [Contact](https://stavrospachoundakis.com/contact) (human check required).

## Elsewhere

- Contact: https://stavrospachoundakis.com/contact (human check required)
- GitHub: https://github.com/stavroschios
- LinkedIn: https://linkedin.com/in/stavrospachoundakis

---

Machine-readable index for this site:

- Full context: https://stavrospachoundakis.com/llms.txt
- Sitemap: https://stavrospachoundakis.com/sitemap.xml
- Developer portal: https://stavrospachoundakis.com/developers

Add `.md` to any page address for a Markdown version.