Four systems aeovi has built for clients. Data sets in the millions, hundreds of websites, countless accounts. Built for volumes where manual work is no longer an option.
URLs as the data set of a web directory.One system, built for millions.
Client project 01 · Directory Scale Engine
34M URL data setEndless scalability, limited only by hardware and rate limits~50 s for 1 or 200 posts5 to 10 s per single job30 pins per account per day24/7 publishingDaily new auction data34M URL data setEndless scalability, limited only by hardware and rate limits~50 s for 1 or 200 posts5 to 10 s per single job30 pins per account per day24/7 publishingDaily new auction data
Overview
Four client projects. No demos.
aeovi built all four for clients. Each one takes on work that would otherwise need an entire department.
Client project 01 / 04Directory Scale EngineRobomotion · AI
Built for millions.
For the relaunch of a web directory, aeovi built a system that reads websites at scale, runs them through a multi-stage filter system and writes directory entries with AI. Designed for a data set of 34 million URLs.
Starting point
The data: 34 million URLs, grown over years and completely overgrown. Casino sites, pill shops, adult content, spam, parked domains, dead links and pages in languages nobody asked for. No descriptions, no maintained records. Not even remotely manageable by hand.
The system
A production line designed for millions. It reads websites in parallel branches, sends every record through a multi-stage filter system and has an AI editorial team write tens of thousands of entries at once.
The crawler
Websites at scale. Read in parallel.
The crawler works in many parallel branches at once. Each branch reserves its own entries, so none gets in another's way. Source code, scripts, ads and navigation clutter become clean, uniform content.
34MURLs data set
The filter system
Filtered in stages. Before it costs.
Not a simple filter but a whole system of stages. The first classifies fetch errors, dead links, redirects and empty pages. The second detects parked domains and wrong languages. The third removes spam, casino, pills and adult content and checks whether the content is fit for a directory entry. Only what passes every stage goes to the AI.
3filter stages before the AI
The AI editorial team
An editorial team that never clocks out.
For every checked entry the AI writes its own directory description. Not one after another, but tens of thousands at once. The system scales endlessly. The only limits are hardware performance and the AI provider's rate limits.
Endlessscalability
The exporter
Done. Straight into the system.
The exporter feeds finished entries straight into the target system or delivers them in packages, whatever the client needs. Every stage keeps its own status. If a run stops, the next one picks up exactly where it left off.
01CrawlerLive schematic
data set 0AI editorial tens of thousands in parallel
Result
34M
URL data set
This is what the system is built for. How many entries the AI ends up writing is a question of budget, not technology.
Endless
scalability
More volume means more parallel processes, not a rebuild. Limited only by hardware performance or the AI provider's rate limits.
0
entries processed twice
Each branch reserves its entries. What one touches, no other touches.
Before
filtered out, not after
The AI only runs on usable data. That keeps API costs under control.
Under the hood
Thought through to the last detail.
With 34 million entries, every edge case shows up thousands of times. That is why half the work sits in details you never see.
Nothing twiceEach branch reserves its entries. What one is working on, no other touches.
Errors classifiedA fetch error does not stop the run. It is classified and handled specifically.
Self-regulatingStuck packages and entries are detected and released again.
Picks up againEvery stage keeps its status. If a run stops, the next one continues right there.
For a client with hundreds of websites, aeovi built a central system that researches, writes, illustrates and publishes dedicated, purpose-built posts for every website. Each post individual, for entirely different projects, all controlled centrally.
Starting point
Hundreds of websites on entirely different topics, all needing their own content all the time. Simultaneously, with as little effort as possible and without logging into each website. By hand that means researching, writing, sourcing images, formatting, publishing. Times hundreds.
The system
One control center for every website. A single job takes 5 to 10 seconds. Then the system researches, writes, illustrates and publishes hundreds of individual posts at once, each for its own website.
The control center
One job. A few seconds.
Set the topic and target website, done. A single job entered by hand takes 5 to 10 seconds. Load whole job lists and each post takes even less. Every website is controlled from here.
5 to 10 sper single job
The editorial team
Researched. Written.
The system researches the topic and writes the post, with everything that belongs to it.
The image studio
Illustrated. Top and in the text.
Every post gets photorealistic images: the header image and the images within the article.
Publishing
Formatted. Published.
The finished post lands formatted in the CMS of the right website. Either as a draft or published directly.
Parallel
1 or 200. Always 50 seconds.
All posts run at once, each for its own website. One or two hundred: after around 50 seconds everything is published. Run time does not grow with volume.
~50 sfor 1 or 200 posts
01JobLive schematic
Control center → websitesPosts 000 / 200
Result
~50s
for 1 or 200 posts
Research, text, images, formatting and publishing. Run time stays the same, no matter how many.
5 to 10s
per single job
Entered by hand. With whole job lists it gets even less per post.
Hundreds
websites from one control center
Each with its own posts, all planned, controlled and monitored in one place.
Complete
with header and in-article images
Photorealistic, generated to fit every post. No image library, no searching.
For a client with countless Pinterest accounts, aeovi built a system that creates content at scale, manages every account centrally and publishes without pause.
Starting point
Pinterest rewards volume and consistency. Countless accounts need fresh pins every day, each with its own content, its own images, its own rhythm. Permanently. By hand, a bottomless pit.
The system
A system that never clocks out. It creates the content, keeps every account in check and publishes around the clock. Nobody posts by hand anymore.
Volume
Endless content. Automatic.
Pins need images, text and destinations, and they need new ones all the time. The system creates them in the volume Pinterest demands.
Accounts
Countless accounts. One control.
Every account gets its own pins. Everything is managed in one place, no handling per account.
Rhythm
Spread across the day. Like a human would.
Pins do not go out in one block but spread across the whole day. Around 30 per account, day after day.
30pins per account per day
Nonstop
At night. On weekends. Always.
Every account stays active and evenly filled. The system keeps running long after the office lights are off.
24/7publishing
01Content at scaleLive schematic
Day 001 · 00:00Pins today 0
Result
30
pins per account per day
Spread across the day, on every single account.
24/7
publishing
At night, on weekends, on holidays. Without a break.
Countless
accounts from one control
Multiple accounts and websites, managed centrally.
0
pins posted by hand
Create, assign, schedule, publish: everything runs on its own.
Source: client project (Pinterest publishing). Simplified schematic, number of accounts symbolic.
For a client in the domain business, aeovi built a system that continuously tracks domain auctions, reconciles them and builds its own valuation basis with history.
Starting point
With expired domains, the valuable data is scattered, fleeting and bundled nowhere. Anyone who wants to value domains checks platforms every day, compares by hand and tracks histories themselves. What was visible yesterday is gone today.
The system
A radar that never looks away. It captures, reconciles and builds a memory, usable in an internal tool for analysis and valuation.
The radar
Scan. Continuously.
Auction sources are checked continuously. No more daily rounds of platforms.
The reconciliation
Capture. Reconcile.
New data is captured, existing entries are updated. Scattered pieces of information become a database.
The memory
Remember. Every day.
Daily data becomes a history. Trends, repetitions and patterns become visible that get lost in manual checks.
The analysis
Analyze. Flag what matters.
Analyses are prepared, anomalies trigger alerts. Decisions start from data, not estimates.
01ScanLive schematic
Radar → historyDay 001
Result
Daily
updated auction data
Captured automatically, without anyone opening a platform.
History
instead of a snapshot
What would be lost without the system stays available as a searchable data point.
Own
data set
Information that is bundled nowhere publicly. Whoever has it earlier decides better.
0
manual spreadsheet upkeep
Platform checks, spreadsheet upkeep and data searches run automatically.
Not every task needs automating. All four projects share the same ingredients. If you recognize two of them in your day to day, a conversation is worth it.
01Frequent.
The more often a process runs, the lower the effort per case.
02Growing.
More cases, more data, more content, without effort growing in step.
03Error-prone.
Less copy and paste, fewer forgotten steps, fewer transfer errors.
04Ties up people.
People take care of exceptions, approvals and real decisions, not routine.
Client project 05still open
The next projectcould be yours.
Where work piles up at scale in your business, we find out together. We look at your processes and work out together what a system for them could look like.
