Groundwork / Assembly

Assemble the system

MapConnectTestRun

AI built on solid ground.

Groundwork makes AI useful where the business actually needs it: the right use cases, the gaps in process and data, the trained people, and the connected systems that give AI something solid to stand on.

Founded by Jacques Tulowitzky · B.B.A. Accounting + Information Systems · Co-author, Current Issues in Auditing

Start a project See the method

The gap

You have the AI.You don’t havethe results.

Getting AI is easy now. Applying it is the hard part. Someone has to decide what it should do, where it runs, who signs off on it, and how it fits the work your team already does.

Skip that, and you pay for activity, not results. Tools stack up, time goes in, and the output still isn’t worth what it costs.

How we apply it

The model has the capability. The setup harnesses it.That setup is the part most teams skip. We don’t.What to build, what to leave alone, who stays in control.Four steps. No shortcuts.

Discover the work

Map the real process first: where AI helps, where it gets in the way, and the permissions, data access, and security boundaries it has to respect.

Architect the solution

Not every problem needs an LLM. Choose each component deliberately, never by default.

Train the team

Build with the people who will use it. The goal is not just a working system. It is shared knowledge, trust, and habits that last.

Hand over ownership

Hand over the tool, the runbook, and the decision rules. The system is yours, and it is not done until your team can run it without us.

The model is only one part.

Human Control

Purpose
Permissions
Approval
Escalation

Models01 / 05
- Generate
- Forecast
- Classify
- Embed
Context02 / 05
- Knowledge
- History
- Skills
- State
Tools & Connections03 / 05
- Search
- MCP
- APIs
- CLIs
Orchestration04 / 05
- Agents
- Workflows
- Routing
- Harness
Reliability05 / 05
- Evals
- Observability
- Guardrails
- Cost

Human control surrounds the system.

What we build

What we build around the AI.

Strategy & architecture

Decide where AI belongs, redesign the work before automating it, and map the models, data, tools, permissions, and costs required to build it well.

Custom systems

Build working AI systems around the real workflow: agents, dashboards, internal apps, automations, routing tools, and approval gates that your team can actually run.

Knowledge & integration

Connect AI to the knowledge and tools your business already depends on: structured memory, databases, search, APIs, MCP, CLIs, and existing applications.

So the system works from your context and can act inside your real environment.

Training & adoption

Train your team to run what we build and get more from the tools already in place. Workshops, hands-on practice, operating rules, and ongoing adoption support.

Field Notes

Real systems, built, tested, handed over.

Five real systems. Identifying details removed. The builds are real.

01In processData + routing

A business-data system

Built: An agentic workspace that maps messy exports into one clean database, then routes every record through about 200 rules in priority order, using a model only for the cases the rules can't settle. Ships with reusable skills and CLI tools, and stays model-agnostic.
Connected: Scattered records from multiple exports unified into one source of truth, with skills that run the repetitive routing.
Controlled: An approval workbook shows a recommendation, a confidence score, the rationale, and a draft for every record. A person accepts, edits, or rejects, and approves every send. Nothing leaves automatically.

Proof of: Knowledge & integration

02Analytics

A business analytics app

Built: A local-first analytics workspace where an owner builds dashboards by dragging widgets together: sales, units, and margin KPIs, channel splits, per-order profit and loss, and top-SKU tables. Modular, so new views can be prompted in as the business grows.
Connected: Multi-source imports unified into one consistent view, with SKUs reconciled across sources.
Controlled: Runs local-first. The data stays on the owner's machine, never fed into an outside model.

Proof of: Custom systems

03Content automation

A marketing content engine

Built: A local tool that harvests trending articles in your niche, ranks them by relevance, and drafts platform-ready posts in your brand voice with the Claude API.
Connected: Pulls from RSS feeds and a news API, then schedules straight to Facebook and Instagram through the Meta Graph API.
Controlled: Nothing publishes on its own. It runs in preview by default, so a person reviews and approves every post before it goes live.

Proof of: Custom systems

04Memory + knowledge graph

A structured knowledge base

Built: Scattered local files turned into a connected knowledge graph the business can see and search. An owned, private memory that lets an owner actually work with the data they have.
Connected: Markdown-linked files joined into one navigable, searchable structure.
Controlled: Files stay local, structured, and under the owner's control.

Proof of: Knowledge & integration

05Multi-agent system

An AI executive team

Built: A suite of role-based agents that works like a C-suite, agentically handling real-time tasks so you stay productive. Built on custom skills and plugins.
Connected: Agents share a common memory and sync on a weekly cadence.
Controlled: Each agent's role and limits are written down, so the team runs as a coordinated system, not a black box.

Proof of: Strategy & architecture

01 / 05

Source note: some examples draw from UW System Small Business AI Center work, independent builds, and continued client work.

Field Notes

Real systems, built, tested, handed over.

Five real systems. Identifying details removed. The builds are real.

Source note: some examples draw from UW System Small Business AI Center work, independent builds, and continued client work.

01 / 05 · A business-data system

01In processData + routing

A business-data system

Built: An agentic workspace that maps messy exports into one clean database, then routes every record through about 200 rules in priority order, using a model only for the cases the rules can't settle. Ships with reusable skills and CLI tools, and stays model-agnostic.
Connected: Scattered records from multiple exports unified into one source of truth, with skills that run the repetitive routing.
Controlled: An approval workbook shows a recommendation, a confidence score, the rationale, and a draft for every record. A person accepts, edits, or rejects, and approves every send. Nothing leaves automatically.

Proof of: Knowledge & integration

02Analytics

A business analytics app

Built: A local-first analytics workspace where an owner builds dashboards by dragging widgets together: sales, units, and margin KPIs, channel splits, per-order profit and loss, and top-SKU tables. Modular, so new views can be prompted in as the business grows.
Connected: Multi-source imports unified into one consistent view, with SKUs reconciled across sources.
Controlled: Runs local-first. The data stays on the owner's machine, never fed into an outside model.

Proof of: Custom systems

03Content automation

A marketing content engine

Built: A local tool that harvests trending articles in your niche, ranks them by relevance, and drafts platform-ready posts in your brand voice with the Claude API.
Connected: Pulls from RSS feeds and a news API, then schedules straight to Facebook and Instagram through the Meta Graph API.
Controlled: Nothing publishes on its own. It runs in preview by default, so a person reviews and approves every post before it goes live.

Proof of: Custom systems

04Memory + knowledge graph

A structured knowledge base

Built: Scattered local files turned into a connected knowledge graph the business can see and search. An owned, private memory that lets an owner actually work with the data they have.
Connected: Markdown-linked files joined into one navigable, searchable structure.
Controlled: Files stay local, structured, and under the owner's control.

Proof of: Knowledge & integration

05Multi-agent system

An AI executive team

Built: A suite of role-based agents that works like a C-suite, agentically handling real-time tasks so you stay productive. Built on custom skills and plugins.
Connected: Agents share a common memory and sync on a weekly cadence.
Controlled: Each agent's role and limits are written down, so the team runs as a coordinated system, not a black box.

Proof of: Strategy & architecture

↓

Ways to Work

Built around how you need to work.

Most engagements combine all three. The balance changes with the project.

Build with you

We work side by side on the decisions and the build. Your team learns the system as it takes shape, not after it is delivered.

Often fits when you want to build capability as you go.

Build for you

We handle the technical build when your team needs a faster or more hands-off path. You still stay involved in the decisions, approvals, and handoff.

Often fits when you need a working system faster, with less technical lift.

Leave you ready

Whether we improve what you already use or build something new, we leave your team with the documentation, operating rules, and practical confidence to keep moving without us.

Applies to every engagement: documentation, operating rules, and the confidence to run it.

01 / 03

Ways to Work

Built around how youneed to work.

Most engagements combine all three. The balance changes with the project.

Build with you

We work side by side on the decisions and the build. Your team learns the system as it takes shape, not after it is delivered.

Often fits when you want to build capability as you go.

Build for you

We handle the technical build when your team needs a faster or more hands-off path. You still stay involved in the decisions, approvals, and handoff.

Often fits when you need a working system faster, with less technical lift.

Leave you ready

Whether we improve what you already use or build something new, we leave your team with the documentation, operating rules, and practical confidence to keep moving without us.

Applies to every engagement: documentation, operating rules, and the confidence to run it.

↓

Who

Meet Jacques.

Jacques Tulowitzky presenting his McNair Scholar research poster

Jacques Tulowitzky presenting the AI Productivity J-Curve at UW-Oshkosh

I work at the intersection of business operations, accounting, and information systems, building AI systems that make sense in the way people actually work.

I work directly with the people who will use what we build. That means learning the real workflow, making the hard decisions together, and training the team to understand the system, not just operate it, so it earns trust instead of being handed over and forgotten.

Connect on LinkedIn

B.B.A. Accounting + Information Systems

University of Wisconsin–Oshkosh. May 2026.

Co-author, Current Issues in Auditing

“Regulatory Audit Innovations in Action: Insights from the Healthcare Industry.”

Founding intern, UW System Small Business AI Center

Spoke publicly for the program, co-built the SOPs, and trained the consultants who came after.

McNair Scholar

Research on Generative AI, Discriminative AI, and RPA in accounting.

Next step

Ready to lay the groundwork together?

The hardest part is starting. We make that part easy. Bring a thought, a problem, or a half-formed idea. If you already have workflows, tools, or plans, bring those too.

The one-hour discovery call is free and is a simple way to get started.

Book a discovery call

Free · 60 minutes · No preparation required

Connect on LinkedIn

AI built on solid ground.