Pricing a deal, replacing an ERP, or putting AI to work — the failure mode is always the same: acting on numbers and processes nobody actually verified.
Based in Prague, Czechia. AI agent implementation, ERP migration, and financial due diligence for Czech & EU small and mid-sized businesses.
About
I spent years as a Senior Auditor inside complex organisations — dissecting workflows, exposing inefficiencies, and writing reports that gathered dust on shelves.
Then I asked a different question: what if we didn't just report the problems — what if we actually fixed them?
Today that splits into three things we do. We run financial due diligence when the numbers need to be independently verified — before a deal, a loan, or a change of ownership. We migrate companies off legacy ERP systems without losing data integrity on the way. And we build AI agents that plug into how a team already works.
Different engagements, one discipline: establish what's actually true before anyone acts on it.
Industries audited from the inside
Maksym Kostenko
Founder · Ex-Senior Auditor · 8 years in finance and audit since 2018 — dissecting how businesses actually run, now verifying the numbers and rebuilding the systems behind them.
Connect on LinkedIn →Marie Kostenková
DevOps & ML/LLM Specialist · Python Developer — builds and maintains the infrastructure the agents actually run on.
Connect on LinkedIn →Due diligence work is exactly that — independent. The findings are the deliverable, with no obligation to build anything afterwards and no incentive for the answer to come out one way.
On migration and automation work, the plan isn't the finish line. We stay on through cutover and go-live rather than handing a slide deck to your already-stretched team.
Services
Three distinct engagements that often follow one another — verify the numbers, replace the system holding them, then automate what's left. Each one stands alone if that's all you need.
A structured, evidence-based review of your company's financial health — built for M&A, lenders, and ownership transitions, not just internal reporting.
Moving you off a legacy or end-of-life ERP onto the right modern system — matched to your specs, data-integrity-first, team trained.
Open-source agent harnesses — OpenClaw, Hermes, or Deep Agents (LangChain) — wired into the channels your team already uses. Two separate choices: which harness fits the job, and which model runs underneath it. Self-hosted when nothing may leave your network, top-tier via OpenRouter when the task genuinely needs it.
Automation amplifies what's already there. We audit and clean your data first — so what you automate is reliable by design.
Capability · part of AI Agent Implementation
We deploy open-weight language models that run entirely on your own infrastructure, with the agent layer on top of them — so nothing, prompts included, leaves your environment. Top-tier models stay available via OpenRouter for the tasks where that trade-off is acceptable. Built for companies where GDPR, client confidentiality, or industry compliance make cloud AI non-negotiable. Your customer data stays in your office, in the EU, under your control.
Own hardware
The most reliable path. We recommend and spec the right machine — a dedicated workstation, server, or even a capable laptop — and handle the full setup. Your model runs locally, always available, zero ongoing API cost.
Private cloud
If on-premise hardware isn't feasible, we can configure a private cloud instance — isolated, access-controlled, and cost-modelled so you know exactly what you're paying before you commit.
Hardware advisory
Not sure what to buy? We advise on GPU specs, RAM, storage, and model selection to match your use case and budget — so you don't over-invest in hardware you don't need or under-spec something that won't perform.
Use Cases
Real engagements across all three services — from an independent report on a live deal to an AI agent running inside your team's chat.
On delivered engagements, research and drafting run roughly five times faster. The point isn't the hours saved on paper — it's that the hours come back to your team. Time that went into searching documents and re-typing the same paragraphs goes into judgment calls, client strategy, and the work that genuinely needs a person.
Around a thousand pages — three novels' worth — all open at the same time. Your full standards library, every court decision, the entire contract archive, in view at once.
The working memory of any human being — you, us, and every expert on your team. Enough for a phone number. Established by Miller in 1956 and unchanged since.
That gap is the entire case for an agent — and its limit. The agent does the holding: it reads everything, forgets nothing, and cites where each answer came from. The person does the deciding, because judgment doesn't improve with more pages in view. That's why we keep a human on every call that carries real risk.
Agents wired into the tools your team already uses.
An agent reachable in Telegram, WhatsApp, or Slack — drafting documents, summarising meetings, researching suppliers, handling back-office tasks. Runs on a self-hosted model when data can't leave the building, or a top-tier one via OpenRouter when the task needs the extra capability.
The harness is a separate decision from the model. For short back-and-forth, OpenClaw or Hermes is enough. For work that runs long — researching a supplier across dozens of sources, or checking a migration file by file — we reach for Deep Agents (LangChain): it plans the task, splits it across sub-agents that each keep their own context, and writes notes to a virtual filesystem, so progress survives a job far bigger than any single context window.
Invoices scanned, data extracted, and entries pushed into your accounting system — cutting manual keying and the errors that come with it. Works with scanned PDFs, email attachments, and supplier portals. On the ordering side, the same agent can generate purchase orders from requisitions, route them for approval, and check order details against inventory and prior purchase history before anything is sent.
An internal agent indexed across your full library of audit standards — ISA and other international standards, 120+ PDFs — so any team member can ask a procedural question and get an answer with the exact standard and paragraph cited, instead of searching documents by hand. Drafts correspondence to colleagues and helps structure audit procedures against the applicable ISA requirements.
Client data is anonymized by a Python script before anything reaches the model — so no personally identifiable or confidential engagement data is ever exposed, even to a hosted model.
Delivered · audit firm
A knowledge-base wiki over international standards, with drafts and client summary reports generated from it. Research and drafting time cut roughly 5×.
Indexed across your supplier and vendor contracts, this agent flags missing clauses, non-standard terms, and expiring agreements before they become a problem — and keeps a running log of what's up to date and what needs attention. Same underlying approach as the audit standards agent, pointed at contracts instead of ISA references.
Pulls spend data across suppliers and expense categories, flags cost-saving opportunities, and surfaces anomalous spend patterns — a duplicate payment, a vendor price creeping up, a category quietly overspending. A natural follow-on after an ERP migration, when the data is finally clean enough to report on reliably.
Indexed across your firm's court decisions and case files, this agent lets any lawyer ask a precedent question and get an answer with the ruling and citation attached, instead of digging through archives. It also keeps a living knowledge-base wiki of past matters and drafts — so a new contract, motion, or memo starts from your own firm's precedent, not a blank page.
Delivered · law firm
A knowledge-base wiki over court decisions, with drafts and client summary reports generated from it. Research and drafting time cut roughly 5×.
Independent verification before a deal, a loan, or a change of ownership.
You're acquiring a company and need the numbers verified before you commit. Ledger and statement review, cash flow quality, hidden liabilities, and a written red-flag report you can take into the negotiation.
You're preparing to sell or raise, and want the problems found by you rather than by the buyer's advisors. We stress-test your financials the way the other side will, so nothing surprises you mid-process.
Getting off a legacy system without losing data integrity.
Your current system is end-of-life, unsupported, or simply outgrown. We plan the cutover, migrate and reconcile the data to audit standards, and stay through go-live — so the opening balances actually tie out.
Before you sign a licence. We map what your processes actually require, pressure-test vendor claims against them, and tell you honestly whether a migration is worth doing at all this year.
How We Work
The method comes from audit practice: evidence first, findings second, and delivery you can hold someone accountable for.
Ledgers, contracts, systems, interviews — whatever the question requires. We document what is actually true before anyone forms a view.
Written up, prioritised, and quantified. Red flags separated from noise, quick wins separated from structural work — so you can decide what deserves funding.
Migration executed, agents deployed, team trained. Scoped and priced per phase, with accountability for every deliverable.
Post-launch reconciliation and performance review. We stay until the outcomes are demonstrable — not merely reported.
Due diligence engagements stop at step 2. The report is the deliverable, it is yours, and there is no expectation that you build anything with us afterwards. That independence is the point — findings shouldn't be shaped by whoever hopes to win the follow-on work.
Due Diligence Sprint
2–3 weeks, fixed fee agreed in advance. Document review, management interviews, and a written findings report. Deal timelines are tight, so the end date is committed up front.
ERP Migration
Starts with a readiness review, then runs by milestone: selection, data migration and reconciliation, cutover, and post-go-live validation. Each phase priced and approved before it begins.
AI Agent Implementation
Scoped as a pilot first — one workflow, one measurable result — before anything is rolled out more widely. No open-ended retainers and no surprise invoices.
If an ERP readiness review or AI pilot leads to the full build, that fee is credited against the implementation. This does not apply to due diligence work — those findings stay independent of whether we ever work together again.
The bottlenecks are no longer what you can build, but what you choose to build.
Process before technology. Signing an ERP licence before understanding your process is how companies end up with expensive systems nobody uses.
AI has a boundary. We apply AI where it saves measurable time — never in places where human judgement and domain knowledge are irreplaceable.
Bad numbers beat no numbers to the decision. Most costly decisions aren't made without data — they're made on data nobody verified. Verification is cheaper than the mistake.
Careful beats fast. The companies that scale don't move faster than everyone else — they move more carefully, with systems that don't break under growth.
AI Boundaries
An agent changes how a decision gets made, not just who — or what — makes it faster. That's a meaningful upgrade for some workflows, and an expensive way to solve a problem a script already handles fine.
Take expense report review. A fixed rules check is a threshold test — over €500 needs a second signature, over €2,000 needs a director. It can't tell you why a contractor's invoice looks off even though every number technically clears policy. An agent can hold that kind of judgment — cross-referencing the vendor's billing history, the contract terms, and the pattern of past submissions before deciding whether something is routine or worth a second look. That's the gap agents are actually built to close — not speed, judgment under ambiguity.
We test every proposed use case against three questions before recommending an agent. Failing all three is a strong signal that a deterministic script will do the job for a fraction of the cost.
Cases with exceptions, edge cases, or context that shifts the right answer — like deciding whether an invoice discrepancy is a typo or a real problem worth escalating.
A ruleset so long and full of exceptions that changing one line risks breaking three others — like an expense policy nobody can fully explain anymore.
Free-text, scanned documents, mixed formats, or anything that requires reading rather than parsing — like reconciling contracts written by five different suppliers.
We run your use case through these three questions before proposing anything. If it doesn't hold up, we'll say so and point you to the cheaper, deterministic fix instead.
Why Us
We blend a background in audit, compliance, and technology — a combination most tech consultants and pure auditors can't offer.
We don't recommend tools until we've mapped your processes. Every recommendation is evidence-based — not vendor-influenced or trend-chasing.
No offshore sub-contractors. No slide decks handed off to your team. We implement alongside you and are accountable for results.
AI where it genuinely reduces costs or saves time — not because it's fashionable. Every automation is justified by a measurable return.
We've audited factories, medical laboratories, fintech companies, retailers, and e-commerce businesses from the inside — finance, operations, compliance, and risk.
You work directly with the expert, not an account manager. Senior-level attention on every project from discovery to final delivery.
Every finding traces back to a document, a figure, or a system you can check yourself. On build work we agree the success metric before starting and close with a before/after comparison.
We use AI where AI earns its place — and a clean Python script where it doesn't. Not every problem needs a large language model, and running one when it isn't justified is just an unnecessary cost. We have deep expertise in how AI agents are priced and how quickly those costs compound at scale. Before recommending any solution, we model the real operational cost and suggest the most efficient path — whether that's a sophisticated AI pipeline or a 50-line script that does the job just as well for a fraction of the price.
Get Started
Book a free 30-minute call — whether you have a deal to verify, an ERP to replace, or a process worth automating. No pitch, no pressure. If it isn't a fit, we'll say so.
No spam. No sales calls. Reply within 24 hours.
Prefer to send details first instead? Fill out the form, or write directly to maxkostenko18@gmail.com.