Michał Skwarczyński

Ventures · architecture · decisions — edition one

00 · The claim

One architect can hold
an entire venture.

In the agent era, one architect can hold an entire venture — capital, product, code. I spent twenty years building those disciplines separately: leading teams of project managers, owning an $8m program budget, managing a board and investors through a funding turnaround. Now they compose. Agents extend what I can build; they don't replace the team I still lead.

Open to senior product & technology leadership roles — remote or EU-based.

One corpusevery fact, onceVenturessix exhibitsAskgrounded assistantColophonhow it works/mcpfor other agents
This site · live
One corpus → every surface
What this isIn one line Six ventures opened up: the constraint each one actually ran into, the decisions taken under it, and what he would do differently now. Not a CV — a record of judgement, with the reasoning left in.
Who it is forReaders Hiring boards and executive teams filling senior product and technology leadership roles, and the CTOs, VPs and product executives who sit on those panels.
How it was builtMethod Agent-orchestrated, under his direction and governed by eval gates — the same operating model used across every venture here. The site is an instance of the argument it makes, not an illustration of it.
How to check itVerifiable One corpus feeds every surface: these pages, the assistant, and the machine endpoints. Ask it something adversarial. It answers from the same facts, and it declines the ones it does not hold.
01 · Why

Twenty years, three disciplines.
They only compose now.

Capital, product and code used to be three careers. Holding all three at once took a founder, a CTO and a PMO — three people who had to agree before anything moved. That was true right up until agents could carry execution.

He collected those disciplines separately and without a plan to arrive here: M&A and corporate strategy, aviation consulting, hedge-fund investment analysis, a fintech he co-founded, an ~$8m MedTech program budget he owned as Chief Solutions Officer. What that adds up to is someone who knows what each part of a venture costs — in capital, in delivery, and in engineering. Agents extend how far that reach goes. They do not replace the team he still leads.

02 · How the work was actually done

Four registers, used precisely. Architected, led, owned, co-built.

Most portfolios flatten every contribution into one verb. These four are the corpus's own vocabulary, applied per venture — including where an agent fleet did the implementation and the decisions were his.

Register 01 · Architected · Steinlog v2 · 2026–present

Cross-tenant leakage made structurally impossible, not merely forbidden.

A bid-coordination SaaS for Dutch public-sector bid teams, re-platformed from scratch after v1 died of Neon compute costs. Tenants are isolated by Durable Object identity: there is no tenant_id column anywhere, so there is no query to get wrong. The cost failure that killed v1 could not be allowed to repeat, and a trust failure would have been existential.

  • Tenant isolation by Durable Object identity — no tenant_id column exists anywhere
  • Corpus promotion blocked on a green eval gate: 84 cases, 60 groundable and 24 adversarial
  • A four-tier market-rail model bans single-market shortcuts by construction
  • Commercial stage stated plainly: pre-revenue, actively prospecting the Dutch market
Steinlog v2 — tenant isolationContractorbid teamappbid roomcollablive editingTenant = Durable Objectno tenant_id column exists1apievent-sourced coreanalytics planeallowlisted, separate2model providersEU-pinned inferencetimestampingRFC-31613
Architected
Read the full record →
Register 02 · Led · MedTech · Chief Solutions Officer

An ~$8m program budget, and three to five project managers running it.

Chief Solutions Officer for a MedTech client, reporting directly to the CEO: owning the program budget, managing and mentoring the project managers under it, and leading preparation of the investor materials for a $20m Round D. Operational discipline and an investor narrative are the same job seen from two ends.

  • Owned an ~$8m program budget, reporting directly to the CEO
  • Managed and mentored 3–5 project managers
  • Led preparation of investor materials for a $20m Round D
  • Introduced the S&OP standard used at Coca-Cola Australia
MedTech — program, compliance, expansionProgram~$8m budget, reports to CEO1S&OP standardCoca-Cola Australia's rigor, imported not grown locally2IoT stackrecurring reliability issues, troubleshot case by case3Regulatory submissionOECD hurdle analysis4IoT vendorfirefought against deadlines, not owned in-houseRound D data room$20m raise, investor-ready5
Led
Read the full record →
Register 03 · Owned · Women-focused superannuation fintech · 2017–2020

A board, an investor group, and a company in serious trouble.

He co-founded a women-focused superannuation fintech, architected its strategy and technology, and raised A$3.5m in angel and seed funding. When the company hit a serious crisis he managed the board and the investors through it and executed a turnaround that enabled the next funding round. One continuous story, not two ventures — and the part worth reading is the middle.

  • Co-founder: owned strategy, technology architecture and product design
  • Raised A$3.5m in angel and seed funding
  • Managed a board and investors through a funding turnaround
  • Recruited a senior team across growth marketing, technology and advertising
Case file — stakeholders through a crisisFoundersstrategy, architecture, product — owned solo1The venturevendors contracted, MVP built end-to-end2InvestorsA$3.5m angel + seed, the runway that had to last3Senior teamgrowth, technology, advertising — recruited to run it4Refinancingcrisis managed, board and investors kept5Key supplierstrustee, managers, administrator — secured under pressure6BoardgovernanceRegulatorregulated financial services
Owned
Read the full record →
Register 04 · Co-built · This site · 2026

The thesis, running. Agents wrote it; the decisions are his.

This site is an exhibit rather than a brochure for one. It is agent-orchestrated on SvelteKit and Cloudflare Workers, carries its own grounded assistant, and is governed by eval gates and rate limits rather than by trust. It also declines to use a vector database — the corpus is small enough that context-stuffing and prompt caching are simply the better call, and knowing when not to reach for retrieval is part of the judgement on display.

  • Agent-orchestrated build, governed by eval gates rather than ad hoc review
  • No vector database and no RAG — context-stuffing plus AI Gateway prompt caching
  • One corpus renders the pages, the assistant and the machine endpoints alike
  • Restricted commercial figures never enter a prompt — enforcement is output-side
This site — corpus to public surfaceCorpustyped facts, one contextAI Gatewayno vector DB — context + cache, not RAG1MCP surfaceevaluateFit: never a verdict, only evidence2Chat UIhuman visitorsEval gatedegraded-mode fallback beats an ungraded answer3Rate limit + Turnstileseparate per-surface budgets, not an open tap4
Co-built
Read the full record →
03 · How to check it

No testimonials. The work is inspectable, so inspect it.

Machine surfaces · read it as an agent
  • /llms.txt
  • /mcp
  • /.well-known/agent-card.json
  • /sitemap.xml
  • /robots.txt
The record · read it as a human
  • Six exhibits
  • Decisions
  • Constraints
  • What he would do differently
The stack under this page
  • Cloudflare Workers
  • Durable Objects
  • R2
  • Workers AI
  • AI Gateway
  • SvelteKit

Where a fact is confidential it is marked confidential rather than blurred into something vaguer — architecture figures are public here, commercial ones are not. Where a claim would be flattering but wrong, the FAQ below says so outright.

04 · The exhibits

Six ventures, with the reasoning left in.

E-01

Steinlog v2

Live · 2026–present · Architected / Owned / Co-built

A bid-coordination SaaS for mid-size Dutch general contractors and specialty subcontractors bidding on public tenders — re-platformed from scratch onto an all-Cloudflare architecture after v1 died of Neon compute costs. Commercial stage: pre-revenue, actively prospecting the Dutch market.

E-02

AI-era builds

Live · 2020–present · Architected / Co-built / Owned

Native iOS/Android apps, a custom e-commerce platform, and graphRAG pipelines, built end-to-end for client ventures under the Dago Digital contract, plus AWS/Vercel/Cloudflare infrastructure work.

E-03

This site

Live · 2026 · Architected / Owned / Co-built

The site itself: an agent-orchestrated build on SvelteKit and Cloudflare Workers, including its own AI chat and fit-evaluation tools, governed by eval gates and rate limits rather than ad hoc trust.

E-04

Women-focused superannuation fintech (anonymized)

Case file · 2017–2020 · Co-built / Led / Owned

Co-founded a women-focused superannuation (pension) fintech: architected the strategy and technology, then steered the company through a serious crisis and executed a turnaround that enabled the next funding round.

E-05

MedTech

Case file · 2020–present · Led / Owned

Acted as Chief Solutions Officer for a MedTech client, owning an ~$8m program budget and leading preparation of investor materials for a $20m Round D.

E-06

Climate investment platform

Case file · 2020–present · Architected / Led

Architected the technology stack and product for a climate-investment platform — vendor selection, a webapp MVP, and an API strategy for administrators, fund managers and data providers. The platform is no longer live; presented here as an architecture case.

05 · The record

The disciplines, and what each one cost to learn.

Twenty years across M&A, aviation consulting, hedge-fund analysis, a fintech he co-founded and an ~$8m MedTech program — the arc that makes the claim above something other than a slogan.

  1. 2005–2007 M&A and corporate strategy — Minova International (now Orica)
  2. 2007–2010 Airline and aviation consulting — Seabury Aviation & Aerospace
  3. 2013–2017 Hedge fund investment analysis — Morphic Asset Management
  4. 2017–2020 Fintech founder
  5. 2020–present MedTech Chief Solutions Officer — Dago Digital
  6. 2020–present Venture architect — Dago Digital
  7. 2026–present AI-native builder — Dago Digital
Program budget owned ~$8m As Chief Solutions Officer for a MedTech client, reporting directly to the CEO.
Project managers led 3–5 Managed and mentored directly — the reason the thesis is not a solo-operator claim.
Angel and seed funding raised A$3.5m For the fintech he co-founded, before managing the board through a turnaround.
06 · Questions

Including the ones with an unflattering answer.

Q-01Does he hold a PMP, PRINCE2, or Scrum certification?
He holds no PMP, PRINCE2, or Scrum Master certification. His project and programme management experience — managing and mentoring 3–5 project managers, and owning an ~$8m MedTech program budget — comes from direct delivery, not a certification body.
Q-02Did he personally write the Steinlog v2 codebase line by line?
Decisions, architecture and governance are his — the tenant-isolation model, the eval gates, the market-rail doctrine. Execution was largely agent-driven under his direction: an AI agent fleet implemented and tested the code against those decisions, the same operating model used across every venture on this site.
Q-03Is Steinlog generating revenue yet?
Not yet. It's pre-revenue, actively prospecting the Dutch market — the live outbound motion into a foreign public-sector market is itself a proof point for international business-development experience.
Q-04Does he use Salesforce or ServiceNow?
At bluepineapple, a software boutique, Salesforce and ServiceNow were tools the firm sold to its clients, not platforms he implemented himself. His own role there was consultative sales.
Q-05What happened with the women-focused fintech he co-founded?
He co-founded a women-focused superannuation fintech, architected its strategy and technology, and raised A$3.5m in angel and seed funding. When the company hit a serious crisis, he managed the board and investors and executed a turnaround that enabled the next funding round — one continuous story, not two ventures.
Q-06Why doesn't this site's AI chat use a vector database or RAG?
The corpus is small enough that context-stuffing plus AI Gateway prompt caching is simply the better engineering call — a vector database would add complexity this content doesn't need. Knowing when not to use RAG is itself part of the judgment on display.
Q-07Is Sweden a live market for Steinlog yet?
Not yet — a Sweden market adapter is in active design, alongside the ongoing Dutch prospecting. The Netherlands remains the beachhead market.
07 · Get in touch

Read the exhibits, then interrogate them.

The assistant answers from the same corpus these pages render, and declines what it does not hold. If you would rather ask a person, the address below reaches one directly.

Ask the assistant