AI & Technology DPVSAIAutomationData EngineProjects ⭐ Featured

DPVS: Software That Runs the Business — Safely

T
TechnoPKG
2026-07-12 📖 4 min read 👁 10 views

Most business software is a filing cabinet. People make the decisions — matching an invoice to a purchase order, posting a cost, reordering stock — and the software writes down what they chose. DPVS is different. It lets software make and act on the routine decisions itself, so a business runs with far less manual back-office work, while people handle only the exceptions.

The hard part isn't getting a computer to decide — modern AI can already read a messy invoice or weigh an ambiguous shipment. The hard part is doing it safely. AI can be confidently wrong, and a business runs on money. A confident mistake, written into the records, is indistinguishable from a fact. DPVS solves this with a simple, strict idea.

The one idea: propose, then check

In DPVS, the AI is never allowed to write to the records directly. It can only propose an action. Every proposal must pass through a gate — an exact, unbreakable set of checks — before it becomes a committed fact. If the numbers don't add up, the gate rejects it. The AI can be as clever as it likes above the gate; below the gate, everything is exact and verified.

The result: a confident mistake is caught, never recorded. You get the speed of automation without betting your records on the AI being right.

See it for yourself

The project page includes live demos. The clearest one shows the safety idea directly. An AI agent is asked to match an invoice to the right purchase order — a genuinely ambiguous task, because vendor names are fuzzy and amounts vary.

Try this: submit an invoice from “Acme Industrial Supply” for 90,000. The AI, matching on the vendor name, confidently picks a purchase order worth 400,000 — the wrong one. The amounts don’t balance, so the gate rejects the posting. The mistake never touches the records. Now submit the same vendor for 400,000: it matches correctly and commits. Same AI, same confidence — the gate is what makes the difference.

What else DPVS does

  • Nothing is ever erased. Every change is a new, signed entry in a tamper-evident chain. The full history is always reconstructible, and altering any past record is immediately detectable.
  • Odd facts are quarantined, not lost. A sensor reading of 999,999 units clearly happened but is clearly wrong. DPVS doesn’t discard it (it’s real) or trust it (it’s wrong) — it sets it aside, flagged and sealed, for a human to review.
  • It protects the business from itself. A system chasing efficiency will pile every order onto the cheapest supplier until one failure sinks it. DPVS enforces human-set limits — if an action would make the business too dependent on one customer or supplier, it’s refused, with a note on the largest amount that would be allowed.
  • Every record answers “why.” Each entry carries who caused it, when, where, and on what basis — so any number traces back to its cause in a single look.

Write in plain language

DPVS has its own small language, GXL, for describing business actions. You don’t have to learn it: the project includes a demo where you type an ordinary sentence and get a valid, ready-to-run action back.

You type: when we ship 250 pumps, record the cost of goods sold You get: event ship {{ param item text param qty count param cost money emit posted account "COGS" debit qty * cost emit posted account "INVENTORY" credit qty * cost }}

Whatever the sentence, the result is always a valid action — the system is built so that an invalid one simply can’t be produced. And because the records are protected by the gate regardless, the business is safe even when the language model is imperfect.

The bottom line

DPVS is a way to let software run the routine work of a business — deciding and acting, not just recording — with a guarantee built into its structure: the machine may be wrong, but it can never be wrong invisibly, irreversibly, or beyond the limits people set for it.

This is a working demonstration of a research system. Every claim above is something you can try on the project page and watch happen.

Tags: DPVSAIAutomationData EngineProjects

Comments (0)

💬
No comments yet. Be the first to share your thoughts!
Sign in to leave a comment.
Table of Contents
Generating...
Share