A working example of institutional AI.
We believe choosing a wine should feel as good as drinking one.
Wine is a useful test of the Conektiq thesis because the problem is not a lack of information. It is turning fragmented science, expert judgment, provenance, market data and changing evidence into something an institution can actually use. We built the knowledge system, scoring capability and governance required to do that at scale — while preserving where every important conclusion came from and keeping human expertise at the center.
Wine knowledge lives across growers, appellations, vintages, chemistry, sensory science, buyers, sommeliers, retailers, critics and consumers. The opportunity was not to add another recommendation algorithm. It was to build enough institutional understanding to make better decisions repeatedly — and preserve what was learned.
Research crossed wine science, earth science, behavioral science and market dynamics. Sources were evaluated by what they could legitimately support. Expert judgment remained distinct from published evidence. The result became durable institutional context rather than a pile of documents.
That context supports a large scoring corpus and the systems required to maintain it. This is the transition Conektiq is built for: understand the institution or domain deeply enough to determine what should become possible, then build and govern the capability that makes it real.
Proof of work · A living institutional corpus
658,005 scored decisions. Each one a distinct judgment — individual vintage, individual terroir, proprietary methodology applied.
The point is not the number of records. It is the ability to keep thousands of judgments useful, traceable and correct as the underlying world changes.
Scale is the easy part. The hard part is that every figure above has to stay true while thousands of changes arrive from books, registries, laboratories, retailers, expert judgment and our own corrections — some of that work done by people, some by machines, over years. Without governed execution, a corpus like this decays quietly and nobody notices until it is wrong in public.
Every source is registered before it may contribute anything, and it contributes at a known standing relative to every other source. A retailer listing and a reference work do not carry equal weight, and neither silently overwrites the other.
An attribute is never a bare fact. It carries its source and its confidence, so a well-attested value and an inferred one are never mistaken for each other — by us, by the scoring engines, or by anyone reading a result.
Work lands in staging, not in the catalogue. It becomes canonical only by passing the gate that stands between the two — checked against what we already hold, and refused when it cannot be reconciled. Completion is not the same as truth.
Every correction is retained as institutional memory, so the same error cannot quietly return through a later load. That record is what lets work continue across years, across people, and across whichever AI models happen to be doing the work this month.
This is what "governed by Sentinel" means in practice. Not a system that works quickly — a system whose output an institution can stand behind.
Every scoring dimension in our system has a foundation in documented science. Wine science. Earth science. Behavioral science. Market dynamics. No dimension exists because it seemed reasonable. Every dimension exists because the primary literature demanded it — and because we could document exactly which research supported it.
We read neurogastronomy. Sensory evaluation science. Behavioral economics. Cognitive psychology. The science of trust and the science of perception. We read the primary sources in every discipline that touches the wine recommendation problem — and we built what the research demanded, not what the industry assumed. We did not build on received wisdom. We built on the papers.
A Court of Master Sommeliers certification takes years of study and carries a pass rate under ten percent. That expertise is not being replaced by a phone — it is being given reach it has never had before. The sommelier's knowledge, encoded in every buying decision and producer story accumulated over a career, is what our system is built to amplify and distribute. We have a standing rule written into our technical guidelines: a score that would embarrass us in front of a knowledgeable wine professional is a scoring failure. Not a defensible output. A failure.
We do not train models. We use inference — asking a trained system a question — which consumes orders of magnitude less energy than model training. We will commit to renewable-energy infrastructure as we grow. We do not scrape or extract data from sources that have not consented. We do not sell user data. The wine community — its growers farming without herbicides, its estates managed for the next generation — deserves a technology partner who holds themselves to equivalent standards. We are trying to be that company.
"The people who care most deeply about wine should be the people who decide how intelligence tools are used in this category. Not the technologists. Not us."
— Conektiq Values Statement
You do not need to arrive with an AI roadmap. Bring us the business problem, opportunity or institutional question. We will do the homework before we ask you to teach us what can be learned from the outside.
We'll be in touch within 48 hours. Thank you for reaching out.