QR

Software Architect, Generative AI

Quentin Roger

AI agent platform in production · 28,000 users · 26 organizations

Scroll to travel ↓

I designed and I build AI Playground, the AI agent platform of Groupe Bouygues: 23,000 users at Bouygues Construction, TF1 and Colas, then around twenty external clients in 2026, each live in under 30 days. I guide the CIOs of the subsidiaries on generative AI and lead a team of three to four people. Before that, ten years delivering Azure and .NET software for CAC 40 companies and SMEs.

The journey · 2010 → today

4

cities: Nantes, Rouen, Marseille, Nice

TDE Transdata

Junior Architect· Jun 2010 — Dec 2012

Joined on a work-study contract during the Supinfo master's degree.

  • —IoT platform collecting GPS and odometric data in public transport
  • —On-board boxes and M2M data collection over 3 to 18 months
  • —Statistical analyses to optimize congested urban networks
  • .NET
  • SQL Server

200,000

personnel payrolls calculated

Ministère de la Défense

Tech Lead· Jan 2013 — Dec 2013

Sopra Steria mission. Improving the performance and stability of the Louvois military payroll software.

  • —Performance and compliance audit
  • —New load balancing system for payroll calculation
  • —Full-year recalculation of the whole population cut from a weekend to a single night
  • —Urbanization of 300 allowance calculations
  • .NET
  • Microservices
  • Oracle

10M

lines of code in the migrated application

Humanis

Tech Lead· 2014 — 2016

Experis IT mission. Migration of an insurance contract management application from Delphi (thick client) to .NET (thin client), for a client with 200 developers and 10 million insured.

  • —Lead developer on the Delphi to .NET migration
  • —Performance audit of the application codebase, with CAST
  • Delphi
  • .NET
  • SQL Server

Smart cards

Edenred and LCL card programs

Gemalto / Numen

Project Manager and developer· 2016

Experis IT mission, after Humanis. Smart card programs for Edenred and LCL: small projects where the project manager both builds and coordinates the outsourced team.

  • —C++ applications driving industrial card printers
  • —Reporting with Jasper
  • —Coordination of an outsourced development team in the Czech Republic
  • —Cryptographic development management
  • C++
  • Java
  • Oracle
  • Jasper

20,000+

referenced parts

Astronics

Project Manager / Tech Lead· Jan 2017 — Sep 2018

  • —Specific solutions for Dynamics AX
  • —Industrial and commercial plan from forecasts and firm orders
  • —PLM reducing product design costs, EN 9100
  • —Over 6 months saved on time-to-market
  • Dynamics AX
  • .NET
  • MongoDB
  • SQL Server

2,700

pre-invoices every month

Tessi

Tech Lead· Jan 2019 — Jun 2022

AZEO mission, having joined in Sep 2018, then Avanade, for Tessi Documents Services.

  • —A €700k project delivered by a team of three
  • —New pre-invoicing automation solution
  • —100% automation from production tool data
  • —Clients invoiced within 5 days of period end, across 40 legal entities
  • Azure
  • .NET
  • Power Platform

200

connected plants worldwide

Eurovia

Technical Architect· Jan 2021 — Jun 2022

AZEO then Avanade mission. E-Cube program: improving the energy efficiency of asphalt plants and reducing their CO2 footprint.

  • —E-drive IoT platform exploiting real-time plant data
  • —Energy consumption and CO2 emissions monitoring
  • —Complete datasets for labs and sales tracking, lower TCO
  • Azure PaaS
  • Azure Functions
  • DevOps
  • ARM Template
  • ADF
  • Azure Digital Twin
  • CodeSys
  • SignalR

1,000+

connected vehicles

Eiffage

Technical Architect· Jan 2022 — Jun 2022

Avanade mission, in parallel with Eurovia and Tessi.

  • —Data platform collecting IoT data from the vehicle fleet
  • —Cross-referenced with IS data for self-service BI
  • —Delivered in 4 months, highly scalable
  • Azure PaaS
  • Stream Analytics
  • Azure Functions
  • ADF
  • SQL Server
  • Log Analytics

10

MVPs delivered, 5 of them in production

Groupe Bouygues

Delivery Manager — Digital Lab technical director· Jul 2022 — Apr 2023

Major digital transformation projects at Bouygues Construction, turning business ideas into concrete products.

  • —Set up and ran a digital lab as technical director, with a team of four
  • —10 MVPs delivered in nine months, 5 of them taken to production, for Bouygues Construction, Bouygues Energy & Services and Bouygues SA
  • —Construction site monitoring, with the Top Site rating
  • —Magic OCR, a reliable OCR for business teams
  • —Owner of the technical choices: Azure services, Power Apps and custom .NET development
  • Azure
  • OpenAI
  • Machine Learning
  • .NET
  • Power Apps

Turning point · Nov 2022

ChatGPT ships

The harness, everything around the model, came one step at a time.

It started with a chat

  1. 01 Chat

    A model you talk to. GPT-3.5, a prompt box, and the feeling that the job just changed.

  2. 02 Chatbots

    Secure enterprise assistants on Azure OpenAI, from 2023. The AI Playground story starts here.

Then, the harness

  1. 03 RAG & tools

    The model reads company documents and calls business tools. First version of the platform, in .NET with Semantic Kernel.

  2. 04 Harness

    Agents, MCP connectors, human approval and guardrails. AI Playground rewritten in TypeScript with coding agents; Claude Code, opencode, pi and Hermes on my desk.

AI Playground

a complete agent harness, designed from scratch

Groupe Bouygues

Software Architect — Building AI Playground· Apr 2023 — 2024

Designed and built AI Playground, the generative AI platform of the group, with a team of three to four people: a complete harness around the models, first written in .NET.

  • —Business agents that automate tasks, multimodal and agentic
  • —AI search engine with RAG over company documents
  • —Connectors to M365, SharePoint and ERPs, through function calling with Semantic Kernel
  • —A wide, vendor-neutral LLM catalogue
  • —Rights managed by group, centralized governance
  • —A team of up to four people, two of them hired by me, technical and product trade-offs
  • Agents
  • Semantic Kernel
  • RAG
  • Azure OpenAI
  • LLM
  • .NET

23,000+

users at Bouygues Construction, TF1 and Colas

Groupe Bouygues

Software Architect — Rollout across the group· 2024 — 2025

AI Playground is rolled out entity by entity: in each one I guide the CIO or the head of data on generative AI. The platform is now sold by Bouygues Telecom Business.

  • —Bouygues Construction: 12,000 users, 7,600 active every month, 2,000+ every day
  • —2.5 million messages and 3,800 agents created by Bouygues Construction users
  • —TF1: 8,000 users
  • —Colas: 3,100 users, close to 2,000 active every month, with ColasGPT
  • —Usage-based business model, which I helped design from the technology and the value delivered: five times cheaper than a per-seat Copilot licence, for a comparable level of service
  • —Connectors migrated from function calling to MCP
  • —Hosting-agnostic: sovereign cloud, France or on-premise
  • —About 200 people reached through generative AI training programs, from novices to experts, and conference talks
  • Azure PaaS
  • Azure OpenAI
  • MCP
  • Sovereign cloud
  • On-premise
  • .NET

23

external clients deployed in 2026, across all industries

Groupe Bouygues

Software Architect — External clients and local AI· 2026

AI Playground leaves the group: the platform is deployed to external clients of Bouygues Telecom Business, across all industries. Alongside it, local AI: models running where the data lives.

  • —23 external clients deployed in 2026, from manufacturing to real estate
  • —Over 5,500 users, 3,600 of them active every month
  • —Each client live in under 30 days: I coordinate the CIO, the CISO, the technical teams and sales
  • —Platform rewritten from .NET and Semantic Kernel to TypeScript in four weeks, in January 2026, with coding agents
  • —10 production instances migrated with no service interruption, 330 test files
  • —Observability in production: cost, latency, errors and telemetry, through an AI gateway (LiteLLM)
  • —Guardrails in production: human approval of sensitive tools, audit logs, token quotas per group, encrypted secrets
  • —Cockpit, a client portal for support tickets, connected through MCP
  • —A team of four for 26 client organizations and over 28,000 users: I keep the core of the platform, the team owns client connectors, support, workshops and pre-sales
  • —Local AI: a playground on iPhone with the model on the device, and inference with vLLM and MLX on RTX 6000 GPUs (OVH), Macs and NVIDIA GB10 (Blackwell)
  • TypeScript
  • SvelteKit
  • Vercel AI SDK
  • PostgreSQL
  • pgvector
  • MCP
  • vLLM
  • MLX
  • iOS

Toolbox

  • Agentic engineering

    Agent harnesses · MCP · RAG · Guardrails · Hooks · Vercel AI SDK · Semantic Kernel · Claude Code · opencode

  • AI

    LLM · Local inference (vLLM, MLX) · Inference optimization · RTX 6000 · Apple silicon · NVIDIA GB10 · Azure AI Services

  • Cloud

    Azure · Sovereign cloud · On-premise

  • Languages

    TypeScript / JavaScript · .NET · C# · Python · PowerShell

  • Build & ship

    PostgreSQL / pgvector · Docker · Git · Azure DevOps · Terraform · Logic Apps · Azure Data Factory

  • Product & interface

    Figma · SvelteKit · Prototyping and MVPs · Product trade-offs

  • Leadership

    Hiring · Team leadership · Pre-sales · Client support · Training and talks

  • Spoken

    French (native) · English (bilingual, TOEIC)

Education

  • Azure Data Engineer Associate · Microsoft Aug 2022
  • Azure Developer Associate · Microsoft Aug 2021
  • Scrum Master · ORCYS 2019
  • Master's Degree · Supinfo 2009 — 2011

Next: a Head of AI or CTO position in a company where the technical lead still builds, owning the AI strategy and the platform, with a team to grow.

Let's talk.

12 missions are a lot to scroll through. Ask the assistant instead.

Ask my resume

Hi! I'm an AI assistant that knows about Quentin's experience and skills. Feel free to ask me anything!

Powered by Qwen 3.8 · always check important information