Parser bots, AI parsers
The bot collects the data you need from open sources, normalises it and delivers it the way you want: a spreadsheet, an API or a message in your messenger. An AI parser brings in a language model that works out how each site is built.
Bot developmentParser bots, AI parsers
What people come to me with
- Data is gathered by hand, copy and paste — several hours a week gone.
- What you need is scattered across a dozen sources in different formats.
- Changes need to reach you immediately: a new price, a new listing, a new posting.
- There are dozens or hundreds of sources, and hand-writing a parser for each one takes too long.
What the work includes
- Collection from sources: websites, open APIs, Telegram channels, files and exports.
- An AI parser built on a large language model (LLM), in two modes: the model extracts the data from the page itself — for rare and varied sources; or it studies the site’s source code and writes a parser for it — from then on the pages are handled by that code, without calling the model.
- Cleaning and normalisation: duplicates, date and number formats, missing values.
- Storage with history — you can see what changed and when.
- Delivery: a spreadsheet, an API, a scheduled report or an alert on the event you care about.
Platforms and technology
Frequently asked questions
Is this legal?
We collect only openly available data, without bypassing logins or protection. Where a source has an official API or terms of use, we follow them. Anything borderline is settled before work starts.
Why should an AI parser write code if the model can read the page anyway?
It can, but every model call costs seconds and money. On a couple of pages the difference is invisible; on thousands of pages from a hundred sites it decides everything. So the model spends its time once per site: it works through the site’s source code and writes a parser for that site specifically. From then on that site’s pages are handled by ready-made code at normal speed, and the whole collection runs far faster than if the model read every page.
If the site changes its markup, does the parser break?
It happens. The bot checks itself: if the source structure changes, it reports the problem instead of loading rubbish into the database. Fixing the parsing is routine maintenance. With an AI parser, the model re-reads the changed site and updates the code.
How many sources can be connected?
There is no limit. Adding a new source is a small separate task that can be done whenever you need it.
How is the price worked out?
I quote the price and timeline once the idea has been analysed and fix them in the specification: as long as the task stays the same, so does the price. The project is split into stages, each paid 50 % up front and 50 % on acceptance.
Do I need a finished specification first?
No. Analysing the idea is a stage in its own right, and the specification comes out of it: what the bot does, on which venue and how it behaves when something fails.
What happens after delivery?
For 3 months I fix, free of charge, any bugs in the work done to the specification. If something critical fails — trading stopping or orders going wrong — I step in within 4 hours, day or night.
Who does the work?
I do, without middlemen or subcontractors: you deal directly with the person writing the code. With some clients I have been working for 3–4 years.
Development starts at 30,000 ₽. The exact price and timeline are quoted after the idea review and fixed in the specification. Pricing policyHow the work runs
Message sent!
Thank you! I’ll reply as soon as I read it.