Skip to main content
Home
Tax & Compliance MinTax

MinTax: an AI document-classification workflow for UK tax compliance

Expense data in, HMRC-rule classification out. MinTax turns a company's expense workbook into a PSA-ready tax analysis, and runs entirely on the user's machine.

MinTax: from manual review to automated PSA classificationBefore MinTax, a company's expense data sat in Excel and a tax adviser went through it line by line against HMRC rules to prepare the PSA, slow and error-prone. MinTax is a desktop app: it validates the uploaded Excel, runs a 14-step analysis that classifies every expense against HMRC thresholds, and stores the results in a local encrypted database. The tax team reviews the classified results and PSA summaries on screen. The manual line-by-line review is gone, and everything runs on the desktop.Where data startsThe MinTax desktop appWhat the team seesCompany expenses (Excel)Manual review in ExcelUpload & validate14-step analysis, HMRC rulesEncrypted local storeTax teamResults & PSA summariesBottleneck removedEvery expense classified to HMRC rulesThe path to followAutomatic data flowRemoved by the engagement
MinTax: from manual review to automated PSA classification
MinTax: system architectureAn expense workbook in Excel is the input. Inside the Electron desktop app, a Python analysis engine runs the 14-step classification and writes results to a local encrypted SQLite store. The SvelteKit interface reads those results, and streams live progress from the engine over an encrypted WebSocket while a run is in flight. The tax team works entirely on screen, and the whole system runs on the desktop with no cloud.InputInside the desktop appWhat the team seesExpense workbook (Excel)Analysis engine (Python)Encrypted SQLite storeTax teamSvelteKit interfaceExcel & PSA exportRuns on the desktop, no cloudThe path to followAutomatic data flowScheduled or async
MinTax: system architecture
MinTax: data flow and live analysisThe tax team uploads an Excel workbook; it is validated on receipt and handed to the engine. The engine runs the 14-step analysis and streams progress to the interface step by step over a WebSocket. Classified results are written to the encrypted store, and the team reviews them and the PSA summaries on screen.UploadedAnalysed step by stepReviewed liveExcel workbookValidate on receipt14-step analysisEncrypted storeLive progressResults & PSAEach of 14 steps streamed liveThe path to followAutomatic data flowScheduled or async
MinTax: data flow and live analysis
Pattern: Manual PSA review automated
Pattern: Runs locally, no cloud

Challenge

UK companies preparing a PAYE Settlement Agreement review employee expenses by hand in Excel against HMRC rules. It is slow, error-prone, and expensive in adviser time, and it gets worse every tax season.

Approach

MinTax is an AI document-classification workflow packaged as a desktop application. It ingests an expense workbook, runs a 14-step analysis that classifies every expense against the relevant HMRC rules, and produces a reviewable result and PSA-ready summaries. It runs entirely on the user's machine, so client tax data never leaves the desktop.

MinTax is a product we co-own with a UK tax adviser, and it is the kind of system Bravko builds: a classification workflow that takes messy, real-world documents in at one end and produces output a professional can act on at the other. Here the documents are expense workbooks, the rules are HMRC’s, and the output is a PSA-ready tax analysis.

The problem. Every UK employer with benefits-in-kind has to decide, each year, which employee expenses are taxable and which are exempt, then settle the tax with HMRC through a PSA. The rules are detailed and full of edge cases. A Christmas party is exempt up to £150 a head, but a penny over and the whole amount becomes taxable. Relocation has an £8,000 exemption. Late-night taxis are exempt only up to a set number of journeys, and only if they are irregular. Long-service awards need twenty years. Done by hand, this means a tax adviser reading a spreadsheet line by line, which is slow, easy to get wrong, and costly.

What we built. A company sets up its profile once: its legal entities and their PAYE references, its annual events, and its policies on things like late-night taxis and relocation. Then, each tax year, it uploads an expense workbook and MinTax runs a 14-step classification pipeline. The pipeline matches each transaction to the right category, applies the relevant HMRC rule, and classifies it as taxable, exempt, or needing a human to look. The result is a set of reviewable summaries and a PSA-ready output, with the working shown so a tax professional can check every decision. That review trail is the difference between automation you can act on and automation you have to double-check by hand.

How it is built. MinTax ships as an Electron desktop app with a SvelteKit interface and a local, encrypted SQLite database. The classification engine is a Python service embedded inside the app. When an analysis runs, its progress streams to the screen step by step, so a long run is legible while it happens. The whole system runs on the user’s machine. There is no cloud, and the company’s tax data never leaves the desktop, which matters to the kind of client that handles other companies’ financial records.

Where it stands. MinTax is built and working. We are taking it to its first users now.

Technologies

  • Electron
  • SvelteKit
  • Svelte 5
  • Tailwind CSS
  • Python
  • FastAPI
  • SQLite
  • WebSocket

Tags

  • AI Workflow
  • Document Classification
  • Tax & Compliance
  • Desktop App

Send us one manual process

Email us the most repetitive data job your team does. We will reply within one working day with how we would automate it, what it costs, and how long it takes. If it is not worth automating, we will tell you that too.

Book a 15-minute call

or write to info@bravko.com