One coherent product
Purpose-bound data checks, a read-only gateway, governed repair, exact readback, and durable proof — as one loop rather than five features.
Home / About
Founder-built · NetherlandsBuilt for the work
after the alert.
An AI system will produce a confident answer from stale, conflicted, or disallowed data without hesitating, because nothing in the pipeline is responsible for asking whether the data should have been used. Refinery exists to be that step.
Why we are building RefineryILLUSTRATION
01
One bad value can travel far.
A support workflow reads the status in its source database. Inconsistent data needs a traceable correction.
Your database customers
| Record | Status | Finding |
|---|---|---|
| #48211 | active | In scope |
| #48213 | enabled | Check value |
| #48214 | unknown | Human review |
02
Make the improvement accountable.
The correction is supported by a rule, approved for one record and verified at the authoritative source.
Before
→enabledProposed
active- Understand the evidence
- Authorize the exact change
- Verify what actually happened
03
Give the next decision better evidence.
A repair receipt explains the change. A separate eligibility check determines what an approved consumer may use.
Source change verified
- Record
- #48213 · status
- Authorized change
- enabled → active
- Fresh source read
- active · matches
- Evidence
- Durable receipt + chain
Why
Data quality should not stop at an alert.
Most data-quality tooling ends where the real work begins. It detects a problem, raises it, and hands a human a queue. Meanwhile the AI workflow downstream keeps reading the same broken record and telling customers things that are not true.
An AI or workflow needs an answer now: trusted, limited, or blocked — with proof when something has been repaired. Not a dashboard, not a score, not a ticket. That is a narrower product than a data-quality platform, and deliberately so.
How it is built
Small enough to move. Strict enough to earn trust.
Refinery is built by one founder. That makes speed and direct feedback possible. It does not remove the need for proof, clear boundaries, or honest readiness claims — if anything it raises the bar, because there is no brand to borrow credibility from.
Meet the founder on LinkedIn. Follow the build on Refinery’s LinkedIn page and X.
One use case at a time
Work directly with design partners to find the source, policy, evidence, and operator requirements that actually matter in production.
One honest status
Say what is proven, what needs certification, and what remains planned — without blurring the three categories to make a page read better.
Being built now
- Purpose-bound data checks for AI and workflows
- A read-only PostgreSQL gateway
- Governed repair with exact readback and receipts
- A focused design-partner model with a real release gate
Deliberately not the pitch
- A platform that already does everything
- An MDM, a catalogue, or a replacement for your source systems
- Unsupervised identity, finance, or broad write decisions
- Scale or outcome claims before customer proof exists
Work directly with the founder.
Describe the approved source, the decision, and what goes wrong. You will get a direct answer about whether Refinery's current PostgreSQL-first scope fits — including a no.