What DataVid does
DataVid is a data and AI consultancy with a strong life-sciences practice. Its accelerator, Datavid Rover, connects legacy systems to large language models, external APIs and internal taxonomies, and its core work is building enterprise knowledge graphs, semantic search layers and GraphRAG systems that ground an LLM in an organisation's own knowledge.
That is genuinely useful work, and worth being precise about. A great deal of value in life sciences is locked inside unstructured and heterogeneous knowledge: clinical and molecular data, regulatory text, scientific literature, internal reports. Making that corpus queryable, connecting it into a graph, and grounding a model on it so answers are explainable and traceable is a real problem, and GraphRAG is a sensible answer to it. Pairing a knowledge graph with retrieval reduces the hallucination you get from naive RAG, because the model is reasoning over connected, curated facts rather than a bag of loosely related text chunks.
So this is not an article about a weak competitor. DataVid does a specific job well: it turns fragmented, multi-format knowledge into something an LLM can use, usually delivered as a consulting engagement around the Rover accelerator. The question is whether that job is the one you actually need done, because the operational questions a regulated business has to answer are a different shape of problem.
Two different shapes of the problem
There are two very different questions hiding under the phrase "AI over our data".
The first is a knowledge question. What does the literature say about this target? Which regulatory clauses apply to this process? Find me the related studies. These are questions over unstructured, heterogeneous text and relationships. Retrieval and knowledge graphs are the right tool, and this is DataVid's home ground.
The second is an operational question. How many serious adverse events by site last quarter? What is enrolment against plan? Which lots are affected if this parameter changes? These are computations over structured, governed, operational records. They do not have a "most relevant passage". They have an exact answer that has to be the same every time, scoped to who is allowed to see it, and provable after the fact.
Most platforms are built for one of these and stretched to cover the other. Integrius is built for the second.
Retrieval and grounding versus deterministic compute
GraphRAG is a real improvement on naive retrieval, but it is still retrieval. It finds relevant facts in a graph and lets a model generate an answer grounded in them. For knowledge questions, that grounding is exactly what you want.
For the operational number, grounding is not enough. A grounded model still generates the figure rather than computing it, and "grounded but generated" is not the same as "calculated". When the answer is a count, a total, a rate, or a trend that someone will act on or defend to an inspector, you do not want the most plausible number. You want the correct one, produced by deterministic computation over the actual records. That is the difference we set out in RAG can't count and, more broadly, in the case for determinism over hope.
Integrius computes. A natural-language question is compiled into a validated query, the data layer aggregates the real records exactly, inside access control and audit, and the model only narrates the already-correct result. Same question, same answer, with its lineage attached.
Product versus project
There is a delivery-model difference that matters as much as the architecture.
DataVid is a consultancy. You engage them, and they build a knowledge-graph and GraphRAG solution for your organisation. That can be the right call when the work is bespoke and exploratory.
Integrius is a self-hosted product. You deploy it on your own infrastructure and own it, the same way you own a database. There is no standing services relationship required to keep it running, and time to a first governed data product is measured in days, not the length of a consulting engagement. For a capability you intend to run for years across a regulated estate, owning the product tends to beat renting the project.
Governance and audit on the serving path
The deepest difference is where governance lives.
A knowledge-graph and retrieval layer is, by design, a knowledge layer. It helps a model find and reason over information. It is not the system that decides whether this user is allowed to see this field, and it is not the system of record for who saw what.
Integrius enforces governance on the serving path itself. Access control is field-level and applied at the moment the data is served, so a consumer who is not entitled to a field does not receive it. Every read and every change is written to a tamper-evident, hash-chained audit trail, and governed actions can require a 21 CFR Part 11 electronic signature with re-authentication and a captured reason. This is active governance: the rules and the runtime are the same system, which is exactly what ALCOA+ and Part 11 ask for. And it runs entirely self-hosted, so no data leaves your network.
When to choose each
Choose a DataVid-style GraphRAG approach when the problem is unstructured, heterogeneous knowledge: literature, molecular and regulatory text, document and graph question answering, grounding a model on a curated corpus of facts. That is real work and they are good at it.
Choose Integrius when the problem is governed, exact, auditable answers over structured operational data, served under enforced access control and audit, on infrastructure you control.
Many regulated organisations genuinely need both: a knowledge layer for the science, and a governed runtime for the operations. They are complementary, not interchangeable. The mistake is using the knowledge-retrieval tool for the operational number, and discovering at audit time that a grounded guess was never going to be good enough.
The bottom line
DataVid makes your unstructured knowledge usable by an LLM. Integrius makes your structured operational data governed, computable and provable, as a product you host yourself.
If you need to find the relevant passage in a mountain of science, that is a retrieval problem. If you need the number to be exact, governed, reproducible and defensible, that is a runtime problem, and it is the one Integrius was built to solve.