Hol a komment?
Collects, every day, when a public figure comments under someone else's post, with a screenshot and a source link.
AI that pays off, not just sounds good.
AI makes money where there is a lot of repetitive text work: documents, emails, quotes, reports. We find those spots in your company, work out what they are worth, and build working systems on them.
Incoming invoice
- Invoice no.
- 2026/118
- Amount
- HUF 486,500
- Next step
- approval, finance
Morning news scan
- Relevant today
- 3 items
- Digest
- ready, with sources
- Sent
- 7:30, by email
Incoming email
- Type
- quote request
- Draft reply
- ready for review
- Routed to
- sales
Scanned archive
- Pages recognised
- 214
- Uncertain spots
- 6, flagged
- Status
- searchable
Question to the assistant
- Answer
- 30 days
- Source
- contract template, section 3
- Check
- one click to the source
Where AI helps
Documents, contracts, invoices
Key fields extracted, summaries, checklists. Incomplete or suspicious items are flagged; people make the call.
Monitoring news, laws, markets
A daily, filtered digest of only the relevant items, each with its source. That is how Hol a komment? and KávéKözlöny work.
Customer questions, quote requests
Classification, draft replies, routing to the right person. The machine writes the boilerplate, you write the personal part.
Scanned archives
Searchable text from old paper, with uncertain spots flagged.
Internal knowledge
An assistant that answers from your own documents and shows where the answer came from.
What you get
- Assessment and roadmapWhere it pays off, what it costs, what the risks are. A concrete pilot proposal with estimated monthly running costs.
- Custom AI automationProcesses built on language models, integrated into your existing systems, measured and capped.
- Document processingText recognition, extraction, classification, and local anonymisation before anything leaves.
- Training and workshopsHands-on AI use on your team's own tasks: good instructions, checking, data protection boundaries.
How we roll it out
Assessment
Interviews and a look at the real material. We find which tasks are worth it and which are not.
Proposal with costsPilot
A working version of one chosen task, on real data, with measurable results.
A tested solutionRollout
Built into daily work, with permissions and team training.
In everyday useMeasure, refine
Accuracy, cost and time saved, measured monthly and improved where needed.
Benefit proven by numbers
What we pay attention to
Minimal AI cost
Our own approach keeps the AI bill as small as it can be: you only pay for what is really worth it.
Every call measured
We measure cost per task and set a daily cap, so the bill holds no surprises.
Sensitive data stays local
Names and identifiers are replaced with placeholders on your own machines before anything leaves. Only anonymous text goes out.
Where it already works
KávéKözlöny
Hungary's official gazette in plain language, by topic, with the date each rule takes effect.
CuliCal
Recipes, calories and weight goals in one place: nutrition worked out from a link, pasted text or a photo of the pack.
Frequently asked questions
Do we need to train our own AI model?
Almost never. Today's language models follow instructions well; we give them your material alongside the question instead of training it in. That is cheaper, easier to change, and the data stays with you.
What happens to our data?
We choose providers and settings where business data is not used for model training. Where that is not enough, we anonymise locally or use a model running on your own machines.
How much does it cost to run?
It depends on the task. Our own approach keeps the AI cost to a minimum, and every proposal comes with an estimated monthly cost.
Will AI replace our staff?
That is not the goal. It takes over the mechanical part: typing, searching, first drafts, comparisons. Decisions and responsibility stay with people, and the system shows what needs checking.
What if the AI is wrong?
We design for it: fixed output structures, source references, flagged uncertain cases, and human approval where mistakes are expensive.
Let's talk about your project.
Tell us briefly where you are, what you need and by when, and we will get back to you.