HELIX · VMD proprietary technology
We do not start enterprise intelligence from zero.
HELIX is the intelligence application VMD owns and builds on: a business semantic model, an operational process model, a deterministic statistics engine and a governed AI layer — configured around one organisation’s systems, entities and governance rather than sold as a standard product.
What VMD owns
A foundation to adapt, not a project to begin.
Most enterprise intelligence work starts by building the same foundations again — a way to describe the business, a way to reach its data, a way to compute something defensible, and a way to keep a language model honest. VMD already owns those, which changes what an engagement is actually about.
Already built
Semantic modelling, process modelling, the statistics engine, the governed AI path, organisation isolation, audit and cost control. This is the part an engagement does not pay to invent.
Shaped per organisation
Which entities matter, what the metrics mean, which processes are modelled, which systems are connected, and what the AI is permitted to do. None of that is generic, so none of it is pre-decided.
Not how it works
There is no signup, no trial and no standard instance. HELIX is not sold as an off-the-shelf reporting tool, and no two deployments contain the same model.
How HELIX understands a business
A schema describes storage, not meaning.
A column called stat_cd means something to the application that wrote it and to nothing else. Two tables holding the same customer agree on nothing, including the key. Most tools skip that problem — they point a model at a database and hope the column names carry enough meaning.
HELIX makes the bridge the product. Two models, both specific to the organisation, both bound to its actual tables and columns.
The semantic model
Entities, their attributes, typed relationships between them, and metrics expressed against them. Joins carry a confidence value and the reason they were proposed.
Tables and columns
the physical schema — names that describe storage, not meaning
Entities, relationships, metrics
modelled per tenant and bound to those tables and columns, with typed relationships and confidence-scored joins
Business context
what a figure means, where it comes from, and what it connects to
The process model
A business process as an ordered set of states, each naming the field and value that puts a record in it. The foundation for an operational digital twin — expressed in the company’s own data rather than in a diagram beside it.
- 01
Received
status = RECEIVED
- 02
In production
status = PRODUCTION
- 03
Ready
status = READY
- 04
Dispatched
status = DISPATCHED
Each state names the attribute and value that puts a record in it, so the model describes the data rather than sitting beside it. Start and end timestamps can be declared per state, which is what makes duration answerable.
Intelligence you can defend
AI explains the finding. It does not invent it.
The finding is arithmetic — z-score, interquartile range, rolling window, linear regression. Same input, same output, every time. Only then does a language model see it, with the evidence attached, and what it writes is checked against a strict schema before anyone reads it. Collapsing those two steps is how software ends up unable to say why it believes anything.
Computed — no model involved
Written — governed by contract
Operational systems
databases and files the business already runs on
Semantic & process model
entities, relationships, metrics, process states
Deterministic computation
anomaly, health, trend
Governed explanation
schema-validated, evidence attached
Decision support
what changed, why, what to look at
Stages one to three are computed. Four and five are written and validated.
No model is involved in the finding.
The model explains it. It does not produce it.
What the governed half runs through
- 01
Context
The finding, its evidence and the tenant’s semantic model.
- 02
Cache
Identical questions are not asked twice.
- 03
Rate limit
Per tenant and per user.
- 04
Local model first
Configurable per tenant — data need not leave the deployment.
- 05
Schema validation
Output that does not match the contract is rejected, not shown.
- 06
Retry, then fallback
One retry, then a configured fallback provider if allowed.
- 07
Audit
Action, timing and outcome recorded against the tenant.
Built for enterprise environments
It learns the business, not the person.
HELIX does not build a profile of how an individual works and does not adapt silently to behaviour. What it accumulates is business context, and a person approves every piece of it before it counts.
Conversation
Someone describes how the business works.
Candidate fact
The assistant proposes a fact, with a confidence value. It is stored as a candidate and used for nothing.
A person decides
Confirmed or rejected explicitly. Nothing is adopted silently.
Company knowledge
Confirmed facts become durable business context and ground later answers.
Data access and control
HELIX reads PostgreSQL directly, discovers its schema and ingests CSV. Further providers are implemented through the same connector layer as an engagement needs them — what is listed here is what runs today, not what is planned.
How VMD engineers integration →Organisation isolation
Row-level security in the database, not only in application code.
Read-only access
Data-source queries are SELECT-only, capped and time-limited by the engine.
Encrypted credentials
Connection secrets are sealed at rest and never returned to a browser.
Rate limits
Set per organisation and per user.
Cost controls
Monthly and daily budgets, with a defined action on limit.
Audit
Assistant actions, configuration changes and data-source runs are all recorded.
What HELIX holds today
Owned, in continuous development.
HELIX is VMD’s technology and VMD keeps building it. The list on the left is what HELIX contains now and what an engagement can be built on; the list on the right is where it is heading and is not built. They are kept apart because a reader deciding whether this fits their environment needs to know which is which.
Current
- Business semantic layer
- Operational process model
- Anomaly detection
- Health scoring
- Trend analysis
- Governed AI explanation
- Human-confirmed company knowledge
- Controlled read-only data access
- Multi-tenant isolation, cost control and audit
Direction — not built
- Process simulation over the operational model
- Richer analytical surfaces
- Additional data connectors
- Approval and verification around an action — being developed once, as platform work in VMD AGENTIC, rather than separately here
- Observe
- Analyse
- Explain
- Recommend
- Approve
- Act
- Verify
The first four exist. Approval, action and verification do not — HELIX does not act on anything today, and nothing in it is autonomous. They are the direction, drawn here so the distinction is not left to interpretation.