Datameal is an AI and data company. We build the platforms, pipelines and governance that make enterprise data usable, then put AI on top of it where that genuinely earns its place — and we develop our own products from the same foundation.
Retrieval systems, document and text pipelines, agent workflows, and the evaluation harness that tells you whether any of it actually works.
Warehouses and lakehouses, ingestion and modelling, quality testing and lineage. The unglamorous layer everything else depends on.
Tools that came out of consulting work and were worth building properly. Developed in-house, in active development.
Fixed scope where scope can be fixed, day rate where discovery honestly has to come first. Each engagement ends with something you own and can run without us.
Systems that use language models for something specific and measurable, rather than a chatbot bolted to a website.
A warehouse or lakehouse stood up properly — infrastructure as code, environments, access model and CI, not a cluster someone clicked together.
New ingestion, or rescuing what already exists. Legacy scripts and scheduled jobs migrated to orchestrated, tested, observable pipelines.
The parts auditors ask about: where personal data sits, who can read it, how long it is kept, and what your AI systems do with it.
AI projects fail at the same place analytics projects do — somewhere between the source system and a trustworthy table. So we start there, every time.
Append-only ingestion with the source payload kept verbatim, so last quarter's answer can always be reproduced.
_ingested_at
Typing, deduplication, history on the dimensions that change, and referential checks before anything downstream reads.
dbt test
One definition per metric, in the language your finance and operations people already use, documented where they will find it.
semantic layer
The same governed tables feed reporting, applications and AI retrieval — so the numbers agree wherever they appear.
one source
Some problems recur on every engagement, and solving them again by hand each time is waste. Those are the ones we develop into products. They are currently in active development and not yet generally available — if a theme below matches something you are wrestling with, we would like to hear about it while the work is still shaping.
Knowing a table is wrong before anyone downstream reads it, rather than after someone presents it.
Measuring whether a model-based feature is improving or quietly regressing, with evidence instead of impressions.
Keeping the record of what data exists, where it goes and who may see it accurate without a quarterly manual audit.
Each phase ends with something you own outright: a document, a running system, a runbook. Nothing is held back for the next invoice.
We map your sources, the numbers people argue about, and what is actually breaking. You get a written architecture and a costed plan — useful even if you then build it yourself.
One week · fixed feeTwo-week increments, each ending with something in production rather than a demo. You see the repository, the tests and the cost dashboard throughout.
Scoped per incrementEither we train your team and leave with the runbook signed off, or we keep it running under a defined support arrangement. Your choice, made at the end rather than the start.
Fixed monthly, or nothing at allA short call is usually enough to tell whether something is a week of work or a quarter. If it is a week, we will say so.
Dublin, Ireland · Working remotely across the EU and UK