Automated SEO/GEO operations that prove themselves
This is not a dashboard that shows charts. It is a scheduled system that measures a website, repairs what is broken, verifies the repair on the live site, and writes down what it did — including what it learned from its own mistakes. It runs unattended across five sites.
Measured on 12 October 2026
171
operational modules
dodaci/*.mjs
42
dedicated guard suites
test-*.mjs
55/55
guard tests green
node dodaci/testovi.mjs
11/11
pages verified on live sites
HTTP 200 → canonical → JSON-LD → llms.txt
705
generated reports
reports/
50
scheduled tasks running
Windows Task Scheduler
How it works — four stages
Input — measurement
Google Search Console and GA4 are queried per site, on the client’s own account. Technical checks: page speed, indexation state, canonical tags, structured data, sitemaps, robots rules, on-page content and internal links. Nothing is estimated — if a number is not measured, the report says “not measured”.
dodaci/analitika.mjs · dodaci/indeks.mjs · dodaci/brzina.mjs
Processing — work and repair
Findings become tasks. Patches are prepared, applied through a git-based deploy chain, and verified on the live site. Every change is preceded by a backup, and a failed post-deploy verification triggers an automatic rollback — the machine does not leave a broken page behind.
dodaci/kod-queue.mjs (backup → snapshot → push → build → verify → revert)
Verification — the part most tools skip
Every published page is checked on the live site through a four-step chain: HTTP 200, correct canonical, JSON-LD structured data, and a link to the machine description (llms.txt). The result is machine-generated and public — not a screenshot in a slide.
dodaci/dokaz-koraka.mjs --svi
Output — report and learning
A daily report is produced and sent. When a mistake happens, the cause is written down together with a rule and a pre-publish check, and a lesson is stored — but only with evidence attached. The machine is not allowed to “learn” something it cannot prove.
data/nauceno.jsonl · data/greske.jsonl · reports/DOKAZ-<date>.md
Where you can verify it yourself
- /proof/ — per-domain evidence, generated by the machine: pages published and verified on the live sites.
- live.braincore.pro/live — the AI agent swarm running right now (WebSocket feed; nodes, board, pheromone trails).
- machine schema — the full architecture, module by module (81 KB, generated from the code).
- snapshot.json — machine-readable state, so another system (or an AI agent) can read it without parsing HTML.
What is not proven yet
I keep technical results and business results apart, because mixing them is how portfolios start lying. The systems above are built, deployed and verified — that part is measured. Organic traffic and AI citations from this machine are still early: Google discovery and model citations move over weeks and months, not days. I will show those numbers when they mean something, not before.
If you want the same measurement-and-repair loop applied to your site, that is exactly what I do for clients — fixed scope, fixed price, written definition of done.