VC Soft / Software house

Software
that
has to
be right.

We design, build and run production software and AI systems for work where a wrong answer costs more than a slow one.

How we build

Most software is built to be fast. Ours is built to be certain: every value is either proven or put in front of a person.

What we build

  1. Product engineering

    Web platforms, back ends and the tools people rely on every day, from the first sketch to the version that runs for years.

    • Go
    • TypeScript
    • React
    • PostgreSQL
  2. Document intelligence

    Pipelines that read scanned paper, cut it into records and hand every uncertain value to a person.

    • JPEG 2000
    • PDF/A
    • XML
  3. AI, used sparingly

    Models where rules cannot see, and rules everywhere else, measured against real data before anyone claims accuracy.

    • Gemini
    • evaluation
    • human review
  4. Infrastructure you own

    Self-hosted in the EU, encrypted off-site backups, audit trails that cannot be rewritten.

    • Docker
    • Caddy
    • Hetzner
    • age

Case study

court

From a shelf of court books to an archive a system can trust.

court.vcsoft.tech
  1. 1

    Receive

    Every file is checked against the book's manifest.

  2. 2

    Prepare

    Each JPEG 2000 master is decoded, measured and rendered; blank pages are found by rule.

  3. 3

    Review

    A person cuts the book into decisions and confirms each field beside its page.

  4. 4

    Deliver

    One XML per book and one PDF/A-2b per decision, each verified before release.

  • 580 pages prepared in under three minutes
  • 36 of 36 PDFs valid PDF/A-2b
  • XML matching the reference output

Measured on a real 580-page book.

Principles

  • Certain, or asked.

    A value is either proven by two independent readings or sent to a person. Never a guess.

  • Every change has a witness.

    Actions are written to a hash-chained log that cannot be edited after the fact.

  • Simple beats clever.

    We build only what the work needs, and we remove what it does not.

  • Measured, not claimed.

    Accuracy is reported from real data, failures included.

Process

  1. 01

    Understand.

    We read your real data before we plan anything.

  2. 02

    Prove.

    A working slice on your own documents, not a slide deck.

  3. 03

    Build.

    Small releases, each one tested, reviewed and deployed.

  4. 04

    Run.

    We operate what we build: monitoring, backups and updates.

Tell us what has to be right.