AI in your firm, built the way accountants need it.
We look through your processes, propose solutions that fit your workflow, build them so every number can be proven — and show your team how to build on them.
We understand your process because we run one.
PTN Accounting is a working practice — bookkeeping, tax, payroll, client questions, deadlines. Most AI vendors learn your process from a discovery call. We already know where the work actually breaks, which steps are arithmetic and which are judgment, and what a reviewer needs to see before signing anything.
Two AI geeks who happen to love accounting.
Nalepa Labs is building this consulting practice in partnership with FutureFy. Between us: years of running tax planning, bookkeeping and tax preparation practices, and years before that inside a regional CPA firm. We have been early adopters for as long as there has been something to adopt — the kind of people who implement the new tool the week it ships, keep what actually works, and throw out the rest. We decided to put that to use helping accounting firms implement AI properly.
- Partners
- Nalepa Labs LLC · FutureFy
- Practice experience
- Tax planning, bookkeeping, tax preparation
- Before that
- Regional CPA firm, South Florida
- Partnership name
- In the making
- Partnership site
- Coming — until then, this page is the front door
Four steps. No slide decks about the future of AI.
- 01
Look through your processes
We sit with the people doing the work and map what actually happens — the intake, the categorization, the reconciliations, the review, the hand-offs. Each step gets sorted: deterministic (code can do it, and prove it) or judgment (a person or a model, with evidence in front of them).
- 02
Propose solutions for your workflow
You get a written proposal that names the steps worth automating, the ones that should stay human, and the ones not worth touching. AI goes where it earns its place. Sometimes the right answer is a script and a checklist, and we'll say so.
- 03
Build, evaluate, secure
Every solution is tested against your real documents before it touches production — measured, not demonstrated. Access is scoped, permissions fail closed, and nothing leaves your environment that shouldn't. Then it is handed over with the evidence and a runbook your team can operate.
- 04
Stay accountable
Agents prepare and prove; your people sign and send. Anything that could be filed, paid, or sent to a client stops at a human gate — and the evidence behind it is one click away for whoever reviews it.
Deterministic where the numbers are. Provable everywhere.
In our industry people rely on the numbers. A tool that is usually right is a liability. These are the rules every solution has to pass — the same rules TransactionFlow was built on.
- Deterministic first
- Arithmetic, matching, lookups and date math run in code, where the answer is checkable. A model gets judgment — never the sum.
- Evaluated, not assumed
- Every process is run against your real documents and measured before it goes live. If it cannot be measured, it does not ship.
- No tolerated hallucination
- A figure with no source does not leave the pipeline. An agent that cannot cite where a number came from reports that it cannot — it does not fill the gap.
- Provenance on every number
- Every value an agent pulls traces back to the statement line, the form box, the ledger entry. The agent defends its number with evidence, not by agreeing with whoever asked.
- Secure by default
- Scoped access, encrypted credentials, fail-closed permissions, tenant isolation. An agent can only reach what the task needs, and every action is logged.
- Human gate
- Nothing is filed, paid, sent or signed by software. People do that, with the evidence in front of them.
Built on, not locked in.
Most software for firms is a subscription to someone else's roadmap: you pay, you get what is in the box, and the box is the limit. What we build for your firm is yours to build on. You know your processes better than anyone, so alongside the solution we show you how to extend it — how to look at a workflow, decide what belongs in code and what belongs with a person, and prove the result — so the next improvement does not have to wait for us.
- 01
The basics first
We introduce the foundation, build the first solution, and verify it works on your real documents. That base matters more than anything layered on top of it.
- 02
Shown how to expand
With the build comes the way of working: how to evaluate a process, where the deterministic line sits, how to check an agent's evidence. Enough to improve it yourself, if you want to.
- 03
Us, when you want us
If you would rather not, or an idea outgrows your team, we are glad to come back and build the next piece. The choice stays with you.
That is how TransactionFlow was built.
TransactionFlow is these rules running in production. Every posted entry traces to a bank line or a statement. A transaction the AI cannot place with confidence is parked and the client is asked — it is never guessed. Reports drill to the transactions behind them. The same discipline is what we bring into your firm, applied to your processes instead of ours.
Have a question, or a process you'd like looked at?
Write to us. We'll talk it through — no pitch, and an honest answer if AI isn't the right tool for that step.
Start the conversation