Agentic AI & Multi-Agent Orchestration
Multi-agent copilots that translate natural language into governed reads and human-approved transactions — 300+ business operations live on Azure and AWS.
the most advanced tools on the market — in production, not on slides
01 · about
From game engines to AI transformation
I help organizations become AI-native — transforming how their products work and how their people perform. As Portfolio AI Lead at Valsoft and Edelweiss Software Group, I lead AI transformation across a global portfolio of vertical software companies, taking agentic AI from strategy to production: governed data access, human-approved actions, and LLMOps telemetry that makes AI ROI measurable.
My daily toolkit: multi-agent orchestration, tool calling, structured outputs, RAG grounding, human-in-the-loop design and AI evaluation. I manage AI as a product portfolio — prioritization matrices, KPI tracking and rigorous ROI evaluation decide what ships.
Before that, I advised government agencies on AI, co-authored winning multi-million-dollar grant proposals, built a Prix Octas 2025 award-winning AI curriculum — and spent a decade shipping large-scale software at Playtika, Ludia, Ubisoft, CAE Healthcare, Dassault Systèmes and Genetec.
02 · method
Transformation, engineered
A five-step operating system I install inside organizations — from first audit to compounding, measured returns. People included, not replaced.
-
Diagnose
Map where AI actually moves your P&L — workflows, data, people. You leave with a prioritized opportunity map.
output: opportunity map -
Prove
A production-grade pilot in weeks, not quarters — with evals and an honest benchmark from day one.
output: measurable pilot -
Platform
Gateway, governance, telemetry: the rails that make AI safe, auditable and affordable at scale.
output: AI rails -
Enable
Train every individual on the most advanced tools on the market — AI-native ways of working, role by role.
output: AI-native team -
Scale
Roll out across products and teams, compound the gains, and keep score with KPIs leadership trusts.
output: compounding ROI
orchestrator --live
One intent, a fleet of agents
This is what work feels like after the transformation — you state the goal, governed agents do the legwork, and humans approve what matters. Replayed here exactly as it runs in production.
pascal@pascalnet:~/transformation — orchestrator
Animated demo: a mission — “close the books 3 days faster” — is dispatched to five governed AI agents that analyze the ERP, build and review a workflow, pass a 98.4% benchmark, and finish with a human approval gate: 3 days of work compressed to 4 hours.
architecture --blueprint
The reference architecture
The governed rails I install: one gateway, one orchestrator, a fleet of specialized agents — every request metered, evaluated and auditable. Watch the packets flow.
03 · expertise
What I deliver
Production AI, not demos — every capability below is live inside enterprise products today.
RAG & Grounding
Retrieval-augmented systems that answer from your data, not vibes: FileSearch stores per ERP, grounded SQL, data validation, and guided next steps operators actually trust.
LLMOps & AI Governance
Centralized LLM gateways that normalize providers and meter tokens, cost and latency per tenant — billing, audit and ROI built in. Eval-gated releases and SOC 2 readiness.
AI Product & Commercialization
From beta rollouts to licensing, installers, demo environments and sales enablement — AI capabilities turned into revenue-generating product lines.
AI-Accelerated Engineering
Orchestrating AI coding agents — Claude Code, Codex, Gemini — through structured, adversarial multi-round reviews. Delivery timelines compressed from months to weeks.
Speaking, Training & AI Talent
AI conference speaker and instructor. Prix Octas 2025 award-winning curriculum, corporate AI training, and hiring engines for AI & Forward Deployed Engineers.
cosine_similarity(skills)
Every skill is a vector
Hover a skill — the others light up by cosine similarity, computed live in your browser. This is exactly how retrieval finds meaning in an embedding space.
live cosine similarity · hand-crafted 8-dimensional embeddings · no GPUs were harmed
04 · trajectory
From pre-training to deployment
A career that reads like a training run — fifteen years of gradient steps, each phase feeding the next.
Deployment — inference at scale
2025 → todayPortfolio AI Lead
Valsoft Corporation · Montréal
- Own the agentic AI reference architecture for a global portfolio of vertical software companies — orchestration, deterministic pipelines, self-correcting planners, eval & benchmark release gates.
- Built the portfolio-wide AI gateway: routing calls to LLM providers, normalizing behavior, and metering tokens, cost and latency per tenant, stage, model and workflow — for billing, audit and AI ROI.
- Govern AI investment with operating-group leaders; run the AI talent engine, screening hundreds of AI Engineer and Forward Deployed Engineer candidates through structured interviews.
Portfolio AI Lead — flagship: Agility Intelligence
Edelweiss Software Group (Valsoft) · Montréal
- Architected and shipped a production multi-agent AI copilot for SAP Business One and Macola: natural language → governed SQL reads, grounded guidance and human-approved ERP writes across 300+ cataloged operations, live multi-tenant on Azure.
- Extended chat into operational automation — admin dashboard, visual designer, worker architecture, triggers and actions that turn AI into repeatable business workflows.
- Quality engineering at AI scale: an honest benchmark harness drove task pass rate from 53% to 98%, while the test suite grew from 145 to 5,500+ green tests with cross-ERP regression pins and CI release gates.
Fine-tuning — the AI pivot
2022 – 2025Project Manager — AI & Programming
CyberCap · Montréal
- Led the AI, programming and cybersecurity curriculum that won the Prix Octas 2025 — and helped launch Initiative Avenir, opening digital careers to the next generation.
- Co-authored a successful government grant proposal securing multi-million-dollar funding; actively supported the merger & consolidation of two organizations.
- Generative AI instructor for youth, corporations and institutions; built chatbots and automation solutions (RAG, OpenCV + LLM pipelines) for clients and government agencies.
Independent AI Consultant
Self-employed
- Tangerine the Monarch Butterfly — an interactive 3D classroom avatar (Unity, Whisper, GPT-4, DALL·E 3) created for a biodiversity conference around the film Butterfly Tale.
- AiTelevision — a 24/7 interactive AI Twitch show: Stable Diffusion, CLIP, text-to-video, cloud TTS and ChatGPT fused inside Unity with a custom real-time chat bot and full content moderation.
- Trained custom Dreambooth Stable Diffusion models on multi-RTX-3090 rigs; taught GPT-4 and prompt engineering to augment professional productivity.
Blockchain & Smart Contract Engineer
Self-employed
- Solidity smart contracts on Ethereum, web3.js integrations, MetaMask support and C# auto-trading systems interfacing with major exchanges.
Training — a decade of shipping
2014 – 2022Unity Developer & Programmer
Playtika — World Series of Poker · Montréal
- Shipped features for WSOP, a top-grossing mobile poker game — gameplay systems, LiveOps skinnable content, localization, monetization — and helped port the C++ engine to Unity C#.
- On Cat Force, helped pivot a single-player match-3 into PvP multiplayer; built Unity level-editor tooling and in-app monetization systems.
Game Programmer
Ludia · Montréal
- C++ gameplay programming on Jurassic World: The Game for Universal Studios — 3D systems, UI, statistical systems and custom export/database tooling across iOS, Android and Windows.
Pre-training — foundations
2005 – 2014Software Engineering Internships
Ubisoft · CAE Healthcare · Dassault Systèmes · Genetec
- AAA debugging tools at Ubisoft · RTSP video for medical simulation at CAE · 3D ergonomics on CATIA V6 at Dassault · PTZ camera protocols for Omnicast at Genetec.
B.Eng., Software Engineering
Concordia University
- Preceded by Computer Engineering at Université de Sherbrooke (2008–2009) and Pure & Applied Sciences at Cégep de Victoriaville (2005–2008).
05 · impact
Measured, like everything I ship
I preach AI ROI for a living — so here are my own evals.
15+
years shipping software
300+
governed operations in production
5,500+
green tests on one AI platform
98
% benchmark pass rate — up from 53%
Prix Octas 2025
Award-winning AI, programming & cybersecurity curriculum (Réseau Action TI).
Multi-million-dollar funding
Winning government grant proposals co-authored for AI & cybersecurity education.
Agility Intelligence
Production agentic AI for ERP/WMS — live, multi-tenant, on Azure.
AI speaker & educator
Conferences, webinars and corporate training across Québec — from first prompt to production.
side quests
The lab
Where new capabilities get stress-tested before they reach the enterprise.
Tangerine, the talking butterfly
Interactive 3D classroom avatar — Unity, Whisper, GPT-4 & DALL·E 3 — built for a biodiversity event around the film Butterfly Tale.
Watch on YouTube ↗AiTelevision — 24/7 AI show
A generative TV channel on Twitch: Stable Diffusion, CLIP, text-to-video, cloud TTS and ChatGPT orchestrated in Unity, steered live by chat.
First-adopter deployments
Agentic vision, code-execution agents, context caching, FileSearch RAG stores — emerging model capabilities, shipped to production first.
roi --estimate
What augmentation is worth
Drag the sliders — a deliberately conservative model of what AI-native ways of working reclaim, before we even talk about new revenue.
directional estimate · assumes 35% automation efficiency on the repetitive share — calibrated on deployed portfolio results
0hours reclaimed / year
0FTE-equivalents freed
$0value unlocked / year
capacity unlocked+10.5%
faq --grounding
Grounded answers
The questions decision-makers ask before an engagement.
What does an AI transformation engagement look like?
A five-step arc — diagnose, prove, platform, enable, scale. You get a prioritized opportunity map in days, a production-grade pilot in weeks, then governed rails and training that make the gains stick.
How fast can we expect measurable ROI?
Pilots ship in weeks with evals and a benchmark from day one, so ROI is measured — not estimated. On my flagship platform, an honest benchmark harness took task pass rate from 53% to 98% while the test suite grew past 5,500 green tests.
Do AI agents replace people?
No — the goal is augmentation. Agents absorb repetitive work under human approval gates, and every individual gets trained on the most advanced AI tools on the market. People included, not replaced.
Which stacks and industries do you cover?
Claude, GPT and Gemini behind one governed gateway, on Azure and AWS — deployed inside vertical software: ERP, WMS, finance and operations. Based in Montréal, working with teams globally in French and English.
How do we start?
Send a message through the contact form with your context and goals. You’ll get a reply within 48 business hours — usually with a first read on where AI moves your P&L.
06 · contact
Let’s build your AI advantage
AI transformation, a speaking engagement, corporate training or a role to discuss — tell me about it.