The documents and policy side lives in DARVIS Docs

DARVIS · DB · structured data

Scattered tables joined into a map of relationships,
asked in plain language

Tables in ERP, MES and SCM are joined through an ontology. Your question becomes SQL, and the answer arrives with the table and column it came from.

Product screen — a conversation demo

Every answer invites the next question

Real use is a drill-down conversation, not a one-shot lookup. Take the answer, then click a follow-up question underneath — every answer arrives with its evidence and the query that ran.

DARVIS DBFictional Food Co. · 2026-05 close · vs last month

Read-OnlyExample — fictional data

Question trail

Keep asking

Contribution margin fell 1.2%p this month — did production get worse, or purchasing?

Here’s how we read your question

Applying the P&L judgement route defined in the ontology — contribution-margin change → materials factor → purchasing vs productivity split

Of the -1.2%p, the materials factor is -0.9%p. Splitting that: purchasing -0.6%p, productivity -0.3%p — purchasing is who to talk to first.

Contribution margin-1.2%pvs last month
Materials factor-0.9%p75% of the change
└ purchasing-0.6%preceived prices
└ productivity-0.3%pyield, costed

Factor	Contribution	Who to talk to
Materials · purchasing (price)	-0.6%p	Purchasing
Materials · productivity (yield)	-0.3%p	Production
Price · discounts	-0.2%p	Sales
Volume · mix	-0.1%p	—

Total -1.2%p — matches the contribution-margin change. Residual 0.0%p.

Evidence

Purchase price master · monthly snapshotCalculated from prices actually received, not contract prices. Multiple suppliers of one item are weighted by received volume.
MES production actuals · yield by processThe drop in good-output rate is costed at input prices and booked as the productivity factor.

SQL executed · 2

SELECT factor, SUM(delta_cm_pp) AS contribution
FROM   cm_bridge
WHERE  period = '2026-05'
GROUP  BY factor
ORDER  BY contribution

Residual 0.0%p — if the split doesn’t add up to the change, we don’t invent an “other” bucket; we don’t answer.

The order of questions here follows the route a real food-manufacturing customer’s corporate-planning team walks when judging P&L: contribution margin → materials (purchasing vs productivity) → variable costs → fixed costs.

The screen and figures are demo data. A real screen is built from your data and your terminology, and where each value came from always comes with it, exactly as above.

Connecting the data isn’t enough to get an answer

PeopleOutside consultants · more headcount

The judgement stays with a person, not the system. When they move on, you start over.

Nothing accumulates

SystemsERP costing modules · BI dashboards

The data piles up, but why it turned out that way lives outside the table.

Nothing gets explained

General-purpose AIBolting a database onto an LLM chatbot

It reads the table names but doesn’t know what those tables mean on the floor.

The context is missing

Judgement that never becomes an asset — this is where cost-reduction efforts keep failing.

What produces the answer is the ontology, not the model

We attach your floor’s words and rules to tables and columns, making one map of relationships. Ask the same question next month and it answers on the same basis.

12
Entities: 42
Nodes: 76
Links: 24
Rules

That is the size of one scenario (as configured in the demo). The next scenario reuses this structure, so only the additions get built.

DARVIS Ontology StudioFictional Manufacturing Co. · inventory and ordering slice · v1.2

Read-OnlyExample — fictional data

SelectConnectAdd entityAdd relation100%

5 entities5 relations3 rules2 sourcesv1.2ValidatedApplied 3 days ago

Click an entity and its attributes, relations and the rules that use it change with it — currently “Inventory”.

One slice of that — the inventory and ordering side — opened up. Click an entity and the rules that read it as a condition appear alongside.

We’re not the only ones saying this

On the public natural-language-to-SQL benchmark (Spider 2.0), the leaders are the approaches that use a knowledge graph.Spider 2.0 · 86%+ · 2026
It treats the semantic layer as essential infrastructure for a data foundation.Gartner · 2025

Where it has been used

Customer	Sector	What we did
Company K	Credit data	Text2SQL · automated reporting PoC
Company L	Food manufacturing	DARVIS phase 1, full contract
Company H	Advanced materials	Smart-factory big-data DARVIS build
Commercestar	Retail & distribution	Database design · data consulting

We don’t publish contract terms or amounts. How a rollout is shaped is something to go through in a conversation.

Your existing systems stay as they are

We connect read-only — nothing is written to your database
We don’t change your ERP or move your data
We go as far as the answer. We don’t control equipment or act on your behalf

Next step

Pick where to start

Month-end cost explained

Read-only connection to your systems
One plant · one product family · one close
Comparison · contribution · evidence
Starts in 4–6 weeks

AX kickoff guide

The three most common false starts
Four conditions for picking the first use case
A 90-day roadmap
A reporting frame for executives, plus a 10-point self-check

Technical review pack

Read-only integration and data flow
On-premise and air-gapped options
TLS 1.3 · AES-256 · SSO/RBAC
An engineer review call if you want one

Scattered tables joined into a map of relationships,asked in plain language