Skip to content

Projects

What has passed through my hands

Client work is described without names and without figures that would allow identification – industry, problem, approach, result. My own projects are named, because they are mine.

Client work

The descriptions deliberately omit names, absolute amounts and dates that would allow identification. Scale is given as an order of magnitude.An illustrative mock-up – the layout of the deliverable on invented data, not a screenshot of client work.

Own projects

Vault – local AI on confidential data

A language model that never sends documents to the cloud

A language model and semantic search running on my own machine (Qwen and bge-m3). Contracts, personal data and trade-secret material are indexed and searched locally – answers come with a source citation and no fragment leaves the computer. This is how I work wherever client data cannot go to an external model.

EquityScopePL

From a register number to a finished valuation report

An application that pulls register data and financial statements, runs them through a valuation engine and assembles a report with a presentation. It exists because three valuations done separately used to mean three different files.

Financial analysis engine

Eight companies, one method

A set of tools and a written procedure: from downloading a filing to a finished workbook, report and presentation. It has its own register of pitfalls – dozens of documented mistakes I do not intend to repeat.

FinGloss

A glossary and search engine for financial terms

176 terms from accounting and corporate finance with full-text search and threshold calculators. The local version works without sending anything outside.

View live

Management presentation engine

A slide is data, not a picture

A presentation generated from an array of data: over a dozen slide types, a bilingual layer, a layout validator and a script-free version for phones and email. It replaced assembling decks by hand.

FinScopePL

From a balance sheet line to the journal entry

Balance sheet and profit and loss analysis with drill-down: the reported line, the accounts that make it up, the entries behind them. It answers “where did this amount come from” without calling the accounting team.

Payroll calculator

Employment contract versus self-employment

A comparison of employment forms under the applicable rules, in a simple and an advanced mode. Written so I could stop recalculating the same thing in a spreadsheet for every conversation.

Ledger controller – AI for an accounting office

A hundred rules instead of the accountant’s eye

The accounting-system export passes through around a hundred control rules (VAT and deadlines, counterparties against the official register, e-invoicing, central-bank rates, account balances) and several hundred tests. A language model suggests a classification for unclear descriptions – the accountant approves. A dashboard for the office owner shows which company failed what today. Runs locally: the office’s client data never leaves its server.