Portfolio ✦ Volume I ✦ Toronto, MMXXVI

Peony Gerochi

I've been building software for 20 years. Most of it in banking and healthcare, where the regulators read your code. I spent two years bringing a digital-health org down from fifty major incidents a year to one. Now I run Zero One Stack and build multi-agent systems that go to production.

Open to
Founding a company · Technical co-founder seat · A few advisory roles
Receipts
50→1 incidents at Loblaw · 70% of devs on AI workflows in a month · Shipped Canada's first PHIPA-compliant GLP-1 chatbot
What I know
Multi-agent systems & LLMOps · Long-running distributed jobs · Building AI under PHIPA, PIPEDA, and banking rules
Where
Toronto · Remote by default · Happy to come in if you're local
Reach me
peonygerochi@gmail.com · LinkedIn · GitHub
~ on the work ~
Peony Gerochi on the Old Bridge in Heidelberg, Germany, with the Neckar river and Neuenheim hillside behind.
FIG. I · Old Bridge, Heidelberg
By the figures
50→1

Major incidents per year at Loblaw Digital. I walked in to fifty. Left two years later with one. It took rebuilding on-call, fixing how we handled incidents, and getting serious about dependencies.

6+yrs

Running consumer healthcare platforms at Loblaw: Shoppers Drug Mart, Pharmaprix, and PC Health. Millions of Canadians on it. PHIPA and PIPEDA the whole way.

1 monorepo · 7 teams

Pulled Shoppers Drug Mart, Pharmaprix, and PC Health onto one multi-tenant platform. Same codebase, seven product teams shipping at the same time without stepping on each other.

60

Engineers across digital health. Grew the team without slowing delivery down.

70%

Of developers using AI in their workflow within 30 days of rollout.

10+yrs

Startup experience, from bootstrapped to large. I've shipped through every stage of that curve.

The executive summary

I build engineering teams that actually ship. Under regulation, at scale, and now with AI doing the heavy lifting.

Peony Gerochi speaking on stage with a microphone, two colleagues beside her at the podium and an audience seated in front.
FIG. II · On stage, mid-talk

At Scotiabank, I took over a 20-person team that wasn't delivering and had them shipping in three sprints, all while keeping the banking systems behind 10,000 daily visits up. At Loblaw Digital, I cut major incidents from fifty a year to one, launched Canada's first PHIPA-compliant GLP-1 chatbot, and got 70% of devs onto AI tooling inside a month. At Zero One Stack, I'm shipping the multi-agent and document systems that turn 12 hours of work into 30 minutes for the teams using them.

I want to do the same thing again, but this time as founder. Find me a company where the engineering is genuinely hard, the regulation is real, and you need someone who can both build the AI and run the team that ships it.

— P.G.

Zero One Stack — current work

Three products. All actually built.

These are what I've been working on. One is in production, one is in build, one is in active architecture. No slides, no mockups.

Verx Plate I
FIG. III · Multi-agent pipeline, in motion Plate
No. 01 · Developer Tooling

Verx

For the monorepo dependencies that keep you up at night.

Sprint dependency work used to take a team 12 hours. Verx does it in under 30 minutes. That's the number that matters, and it ships today.

It's a multi-agent pipeline. The agents analyze dependencies, plan upgrades, validate them, and open the merge requests for you. Everything runs in sandboxed Cloud Run with verification, rollback, and retry loops for when something fails. Blast-radius analysis catches the downstream package and peer conflicts before they hit prod. Works natively with Nx, Turborepo, and pnpm.

Stack TypeScript · Claude API · Google Cloud Run · Node.js · Postgres · Nx
Sertus Plate II
FIG. IV · Document processing at length Plate
No. 02 · Public Sector AI

Sertus

Turns Canadian public-sector agreements into datasets you can actually query.

Built for the documents that break everyone else's pipeline. Two-hundred-page collective agreements, parsed without losing fidelity. The interesting part is the architecture: a checkpoint-resumable workflow with a parking mechanism I built to handle serverless timeouts. If a job dies mid-stage, it picks up where it left off.

Model routing keeps costs sane. Frontier models when the reasoning actually needs them, cheaper models everywhere else. Distributed workers scale horizontally so it can handle real volume. You can't fake this with a Pinecone demo.

Stack Next.js · Claude API · Docling · GCP · Cloud Run · pgvector · Postgres
Omera Plate III
FIG. V · Persona, after-hours Plate
No. 03 · Consumer · In Build

Omera

AI personas for people who shouldn't have to learn to prompt.

Here's my bet: most people will never learn to prompt, and they shouldn't have to. Omera is consumer AI where the persona handles the prompting for you. $5 a month, one app, built for the user who wants to open it every day and have it just work.

PRD's done, the product vision is locked, and the architecture is in motion on Claude Managed Agents. It's the first Zero One Stack product aimed at regular people, and the first one I'm pricing as a consumer product instead of hiding behind enterprise margins.

Stack Claude Managed Agents · Next.js · Cloud Run · Postgres
Career, in chapters
Apr 2025 — Present Toronto

Founder · CTO

Zero One Stack — AI Products & Infrastructure

Founded and run a studio building AI systems for regulated and enterprise customers. Three products in flight: a multi-agent platform that cuts sprint dependency work from 12 hours to 30 minutes, and a document-intelligence system that handles 200-page collective agreements without losing detail.

I own all of it — architecture, hiring, AI infrastructure, operating model, pricing, product. The same work I'd do as your founder, except I'm doing it now.

Multi-agentCloud RunLong-running jobsHybrid model routing
Jan 2023 — Mar 2025 Loblaw Digital

Senior Software Engineering Manager · Digital Health

Healthcare platforms under PHIPA & PIPEDA — 60 engineers

Took over a digital-health engineering org running fifty major incidents a year. Two years later it was running one. To get there I rebuilt how we did incidents, fixed on-call, cleaned up the dependency mess, and got delivery into a cadence product could actually plan against.

While that was happening, I shipped Canada's first PHIPA-compliant AI GLP-1 chatbot for Shoppers Drug Mart, and got 70% of developers using AI in their day-to-day inside the first month. Most orgs take a year to do that.

50→1 incidentsRegulated AIAI adoptionOperating modelHealthcare
Sep 2021 — Jan 2023 Loblaw Digital

Software Engineering Manager · Digital Health

Consumer healthcare platforms — 25 engineers

This was the groundwork for the turnaround that came after. Built the basics the org needed to run: predictable delivery, dependency management, real hiring and onboarding. None of it shows up in the headline numbers, but the headline numbers don't happen without it.

Operating modelHiring systemsGCPFirestore
Feb 2020 — Sep 2021 Scotiabank

Principal Software Engineer

Three teams · seven applications · twenty reports

Walked into an engineering org that wasn't delivering. Three sprints in, it was. Same playbook I'd later run at Loblaw. Kept the banking systems behind 10,000 daily visits running while I did it, and modernized the legacy architecture without breaking the regulatory or security constraints.

TurnaroundReliabilityJavaBanking regulation
2010 — 2021 Consultancy

Founder & Technical Lead

ZeroOneStack Consultancy — first time around the founder loop

Eleven years on the side of the day job, advising founders from pre-seed through Series A. Built platforms for Diversio, Trulioo, Higher Ground Education, Strata, and others. This is why I know going founder now isn't a leap — it's the next thing.

Founder advisoryFintechEdtechE-commerce
“
AI can help you build the product. It can't build the team that ships it. That part is still on the operator. That's what I do.

I build engineering teams that work. Teams that ship under regulation. Teams that hold the line on reliability. Teams that finish the hard stuff when the founder isn't around. AI makes those teams faster. It doesn't replace them.

Peony Gerochi · Operating principle
Group photograph of the Loblaw Digital health team in their Toronto office, posed in front of a large screen.
FIG. VI · The team, on the day Loblaw Digital · Toronto
The practice

What I use, and what I think about it after shipping with it.

No. 01 — Intelligence

AI systems & agent platforms

Multi-agent orchestration, RAG pipelines, LLMOps, Claude API, Anthropic Agent SDK, LangGraph, hybrid model routing, prompt versioning, agent permission systems, MCP & A2A protocols, retry & fix loops, inference economics
No. 02 — Evals & AgentOps

Reliability for non-deterministic systems

Regression-grade evals, LLM-as-judge pipelines, AI observability, hallucination detection, AI safety guardrails, runtime governance, agent incident containment, token-level cost attribution, human-in-the-loop escalation, audit logs & BYOK
No. 03 — Architecture

Platforms & distributed systems

Cloud-native architecture, long-running job orchestration, event-driven systems, checkpoint-resumable workflows, monorepo architecture, Module Federation, fault-tolerant pipelines, API gateway design, resiliency patterns, infrastructure modernization
No. 04 — Tongues

Languages & frameworks

TypeScript, Python, NestJS, Next.js, React, Node.js, Java, React Native, Nx, Turborepo, PHP, Shopify, WordPress
No. 05 — Ground

Cloud, data & messaging

Google Cloud Platform, Cloud Run, Postgres & pgvector, Pinecone, Qdrant, Weaviate, Cloudflare R2, GKE, Pub/Sub, Kafka, RabbitMQ, Firestore, Apigee, AWS, MongoDB
No. 06 — Operating model

Leadership & org design

AI-native team design, Hiring & onboarding systems, Incident response, regulated delivery (PHIPA, PIPEDA, banking), engineer:PM ratio design, retention & succession, cross-functional product/design partnership, board & exec reporting, technical due diligence
Correspondence

Need a founder who's done this before? I have. Let's talk.

I'm open to founder conversations, technical co-founder seats where you actually mean it, and a few advisory engagements. The harder the constraints, the more interested I am. Be specific in your note and I'll do the same.

Send a letter → View GitHub
By Post
peonygerochi@gmail.com
In Person
Toronto, Ontario