NNMQWeb Systems
Belgrade at night — city and river panorama

NMQ

Senior AI / automation engineer · Belgrade

26 products. 2.49M lines of code.Ready systems, hourly work or a full-time role.

12 live domains, 3 finished installers and 5 production systems with measured proof — 55/55 guard tests, 277/281 swarm tests, 1,238 blocked actions in 7 days. Asynchronous, written, no status calls — you choose how we work.

Write to meSee the projects

Work · Braincore, BeoPro, ACC…

See the work

Works

Live projects — Braincore, ACC, BeoPro…

Open

Capabilities

Architecture, GEO, calculator, maintenance.

Open
The brain of the site

We do not build a website. We build a brain that learns about your company.

A classic website is a shop window. Ours is an employee: an assistant that knows your price list, answers customers 24/7, and is smarter every day than it was the day before.

01

We fill it with your knowledge

Services, price list, documentation, frequently asked questions — everything your team knows, the assistant learns from your sources. It is not a generic bot from the internet.

02

It learns about you — and remembers

Every visitor inquiry becomes part of the knowledge base. Over time the assistant understands better what your customers are looking for and how to lead them to a decision.

03

It qualifies before contacting you

Instead of an empty contact form, you get a ready inquiry: who, what they want, what budget, and what problem needs solving. You simply reply.

04

An AI architect connects everything

Behind it stands NMQ.Architect — design and development built to measure, plus monthly knowledge base maintenance so the brain always knows the latest.

Every assistant answer goes through your knowledge base — it never invents prices or services. If it does not know, it honestly points to email. That is the difference between an “AI chatbot” and a brain that works for you.

Request an analysis
Four machines

Four machines that work instead of you

Along with your site you get the machines that guard it and push it forward: one watches whether Google and AI see you, one weaves your AI models into your own work, one keeps you from damage, one prepares your knowledge for AI. All of them run on your own machine — no monthly fees for someone else’s tools.

Check your site for freeTry the machines below

Visibility 360

01

Who it is for: For a site owner who does not know whether people find him, and has no time to check pages, speed and AI mentions every day.

How it works
  1. 1Walks your sites — every page, alone, at 05:10
  2. 2Measures — index, queries, speed, markup
  3. 3Finds faults — duplicates, empty descriptions, pages Google cannot see
  4. 4Prepares fixes — ready changes, verified so nothing breaks
  5. 5Writes the report and the proof — numbers and a SHA-256 signature
What you get
  • ✓Daily report: what was found, what was fixed, what waits for you
  • ✓A page the client opens alone and sees his own numbers
  • ✓Proof in a signed file — the report cannot be edited afterwards

Our case: 5 sites · 187 pages · daily run without a single human

Star

02

Who it is for: For someone who has several AI models on his machine and feels none of them understands him — because none knows exactly how he works.

How it works
  1. 1Lists your models — what each is good at, how much memory it needs
  2. 2Breaks down your work — pulls jobs and rules out of your description
  3. 3Assigns the best model — for each job, the one that really does it
  4. 4Builds the chain — who works first, who second, what passes on
  5. 5Measures understanding — score 0–1 with proof and what is missing
What you get
  • ✓Model passport — one document describing your work and who does it
  • ✓Understanding score 0–1 with proof (a number, not a feeling)
  • ✓Everything stays on your machine — no cloud, no subscription

Our case: understanding 1.0 · 7/7 jobs covered · 7 models in the chain

AgentGuard

03

Who it is for: For anyone letting AI actually do things — send, change, delete or pay. Existing tools only record the damage afterwards; this one stops it before.

How it works
  1. 1The agent asks for an action — "send a 1,500 refund"
  2. 2Rules check — limits, allow-list, sensitive actions
  3. 3Decision — allow · ask a human · STOP (unknown is always STOP)
  4. 4Execute and verify — if verification fails, state is rolled back
  5. 5Writes everything — a permanent log: who, what, when and why
What you get
  • ✓The damage that never happened — a result no statistic shows
  • ✓A log of every decision, with the reason
  • ✓A management report: how many actions, how many stopped

Our case: 7/7 scenarios · a forced failure was fully rolled back

Context Forge

04

Who it is for: For a company feeding AI its prices, contracts and procedures — and not wanting AI to quote a 2023 price list or the wrong clause.

How it works
  1. 1Ingests documents — prices, procedures, contracts, notes
  2. 2Removes duplicates — the same content stored twice stays once
  3. 3Flags contradictions — they go to a human, not to AI
  4. 4Measures freshness — every fact gets a date and age score
  5. 5Emits the knowledge pack — source, owner, date and signature
What you get
  • ✓A knowledge pack ready for your AI tool (plus a human version)
  • ✓A list of duplicates, contradictions and outdated facts
  • ✓Every chunk carries its source and a SHA-256 signature

Our case: 7/7 situations detected · a pack is built in 0.15 s

Try it, no sign-up

Try the machines — free, inside your browser

Everything below is computed on your own machine: no answer is sent anywhere, there are no paid calls and no waiting. This is a shortened demo — the real machines do much more.

Tick what is true for your site — the score is computed instantly.

This is a simplified demo. The real machine does more and leaves proof in a file (SHA-256).

Machines · Paketi

Five machines we build and run.

Pet mašina koje pravimo i vodimo — bez cena na ovoj strani, cene idu na razgovoru. Jedna od njih je naš AI Ops proizvod: odobrenja, audit, evali i izveštaji.

01

Visibility 360

Daily checks whether Google and AI see the site; signed report.

02

Star

Chooses the right local model for each job and measures understanding.

03

AgentGuard

Stops a dangerous action before it happens; logs every decision.

04

Context Forge

Turns prices, contracts and procedures into a signed knowledge pack.

05 · Our product

AI Ops machine

Policy per action, signed approvals, hash-chained audit, evals, SLO and reports. Naša mašina: odobrenja, audit sa lancem, evali i izveštaji.

See the machine →
+

Something else?

If the job is a machine, we build it — job engine, lead machine, invoice and collections desks.

In practice

What it looks like in practice

Screenshots of the real systems — none of them is a mockup. Click an image to enlarge.

Captured from the live domains on 17 October 2026.

AI integrations

What we can build in

These are modules we can build into your site. Some already run in our own systems — Visibility 360, Star, AgentGuard, Context Forge and RADAR. Others we build per project, through external services (WhatsApp, HeyGen, Stripe).

AI assistant 24/7

Knows your prices and services, answers customers at any hour and leads them to a decision.

DeepSeek · RAG

Automatic messaging

WhatsApp messages that close the sale without your involvement.

WhatsApp Closer

Finds opportunities

Scouts jobs and ranks them by how good a fit they are for you.

Sniper

Tracks AI visibility

Whether ChatGPT, Perplexity and Gemini cite you — score 0–100.

RADAR

Competitor analysis

Where competitors win, where the gaps are, and how to take them over.

War Room

Automatic video

15-second Reels from text — in several languages, without filming.

HeyGen

Automatic billing

Upsell and billing that moves to a higher tier on its own.

Stripe

Remembers and learns

Every inquiry becomes part of the knowledge base — the system is smarter every day.

Memory · RAG
Featured work · Braincore Pro

Braincore Pro — a decentralized AI swarm

A swarm of specialized AI agents inside your company. There is no master node that can fail: the nodes find each other on their own, they share a single board, and the work is picked up by whichever one is the most available.

< 2 s

node discovery (42 ms across 3 nodes)

277 / 281

tests passing (3 known failures being fixed)

0

npm packages — Node built-ins only

24 / 24

golden eval on a real model

No central orchestrator

A task handed to any node is executed by whichever node is freest. Proven in the test suite, not on a slide.

Survives a hard kill

Under SIGKILL and an hour of load: 0 lost, 0 double-executed tasks, node death detected in ~1.3 s.

Encrypted peer traffic

HMAC-signed frames and AES-256-GCM encryption. No libp2p, no brokers, no Zookeeper.

Only numbers travel

The registry accepts eight fitness fields and rejects content — we cannot leak data we never receive.

Technology

  • UDP gossip over dgram — probes every 300 ms, fanout 2
  • CRDT blackboard (last-write-wins + vector clocks)
  • Redis queue with a zero-dependency RESP client
  • Stigmergy: pheromone trails, 30 s TTL, true exponential decay
  • RSI levels R0–R5 — every change needs human approval

All figures come from braincore.pro — from their test suite, not from marketing.

Visit braincore.proWatch the live swarm
Featured work · AI Command Center

AI Command Center — three machines, pick one

You are not buying a retainer, you are buying a machine. You pay, you get a key, you open START and you pick the machine that does your job: SEO+GEO, Analytics Board or Architect+Commander.

$149 / $249 / $499

STARTER / GROWTH / OVERLORD — per month, per machine

3

machines on offer — you pick one

$19

Mirror Clone Report (optional, one-time)

9

AI probes in the Mirror report

Machine 01 — SEO + GEO

Daily citation and AI Overviews checks, Maps readability, ASOV tracking, citation locks, and auto-fix for FAQ and ghost pages.

Machine 02 — Analytics Board

Drop → mart ≤2400 → push ≤600 → four packages. The same key model, no retainer.

Machine 03 — Architect + AI Commander

A merged Architect license and Commander workbench in one package.

Mirror — proof against clones

Nine AI probes against your real site: real vs fake clone, citation fragment score, and agent readiness.

How it works

  • Payment via Creem (merchant of record), cancel anytime
  • 7-day money-back on monthly plans
  • Prices are machine-readable (llms.txt) — so AI does not invent wrong numbers
  • Key → START → machine; no retainers, no meetings
  • Desks: /seo-geo/, /board/, /commander/, /mirror/

Prices and plans come from aicommandcenter.pro — $149 / $249 / $499 per machine (USD, product pricing). NMQ services on this site are quoted in EUR. Test suite status: 277 of 281 passing; 3 known failures are listed openly in the project page.

Visit aicommandcenter.proOpen the live Board
Work

What we have already shipped live

Live systems — AI, real estate, auto, casino, tracker. No price list here. If a project interests you, write by email.

Contact

Interested in a project?

Fill in three fields and the message lands straight in our inbox. No mail app, no waiting.

Curious about a project?dejanswat@gmail.com

Custom builds

We build software, fix bugs, and make whatever you need.

We are open for cooperation — from a small fix to a complete system or a machine made to your needs. Tell us what you need, we build it and maintain it.

Custom software

Desktop, web, or server software — from idea and specification to delivery.

Bug fixing & maintenance

Existing system misbehaving? We find the root cause and fix it, without rewriting from scratch.

Automation & machines

Systems that run themselves: robots, agents, integrations, and data flows with no manual work.

Cooperation & projects

Open to joint projects, white-label work, and long-term partnership.

Send a request/provera
Four ways to work together

How we can work

All four options are open and all are asynchronous — written communication, no status calls, decisions are documented. Pick what you need; no amount is on this page because terms are agreed directly.

Full-time role

Salary, remote, asynchronous. I work as part of your team — written communication, no daily meetings.

Hourly work

A trial period or a smaller scope — you pay only for what is used. The natural way to see how I work before a bigger commitment.

Ready-made system

A system that already runs: SEO Operator OS, NMQ Boom, iGaming Suite, RADAR, AgentGuard. Licence or handover — code and documentation come with it.

Built to order

Tell me what you need — I build it from scratch, with tests, proof and public code wherever possible.

  • 26 products, 2,486,677 lines of code, 39,742 files
  • 12 live domains and 3 finished installers (.exe)
  • 5 systems in production with measured proof (55/55, 277/281, 1,238 blocked actions)
  • All solo — from architecture to deploy and monitoring
  • Privacy by design: local models, air-gapped without internet, data never leaves the machine
See the codeWrite to me

This page has no price list. Terms are agreed directly — write by email.

Frequently asked questions

Briefly, before we start

Prices for Serbia: Core €1,200–2,200, Business €2,500–4,500, Enterprise from €5,500. The quote is fixed before we start. For international clients — a different figure by email.