Organizing Brownfield Data Across Multiple Plants.
Kobai + Databricks Genie: Building a Fully Contextualized Chat Environment on Your Lakehouse
Kobai + Databricks Genie: Building a Fully Contextualized Chat Environment on Your Lakehouse
14:40
Databricks Genie is a powerful conversational interface. Kobai is the semantic context that makes it accurate, consistent, and scalable across the enterprise. This post explains how they work together and what changes when Genie has a shared semantic model behind it.
Databricks Genie makes it possible for business users to ask questions about Lakehouse data in natural language. It is a genuinely useful capability as it lowers the barrier to data access and reduces the dependency on analyst teams for routine data questions.
But as organizations try to scale Genie beyond a single team or data domain, a consistent challenge emerges. Each Genie space develops its own business logic. Definitions drift: “revenue” means one thing in the finance space and something slightly different in the sales space. Cross-domain questions — those that require connecting customer data with operational data, or asset data with maintenance history — break down because there is no shared semantic foundation.
This is not a Genie limitation. It is a context problem. Genie is a natural language interface; it answers questions from the data it can see and the business logic defined in its space. The challenge of scaling it is the challenge of making that context shared, governed, and consistent across the enterprise.
|
Databricks Genie = natural language interface to your Lakehouse data. Kobai = the shared semantic context that makes Genie accurate, consistent, and enterprise-wide. |
The Genie scaling problem
Genie works well for a defined, single-domain use case: a sales team querying CRM data, an operations team querying maintenance records, a finance team querying revenue tables. In each case, the business logic is specific to the domain and the team, and it can be configured directly in the Genie space.
The challenge appears when an organization wants to do two things with Genie:
- Scale Genie across multiple business units, so different teams can ask questions about different domains from the same Lakehouse
- Enable cross-domain questions — questions that span operations and finance, or customers and supply chain, or assets and workforce — that require connected context to answer correctly
Both of these require a shared semantic foundation. Without one, scaling Genie produces a proliferation of spaces, each with its own business logic, each answering the same concepts differently. When a user asks a cross-domain question, the context boundary of a single space is not enough to answer it.
|
As Genie scales... |
The problem that emerges |
The consequence |
|
Across business units |
Each space defines its own business logic. Terms like “customer” and “active” mean different things in different spaces. |
Answers are inconsistent. Users in different teams get different answers to the same question. |
|
Across data domains |
Each space is bounded by its own data context. A maintenance team’s space cannot access the operational scheduling data in a separate space. |
Cross-domain questions cannot be answered. Users revert to manual investigation. |
|
Across the enterprise |
Configuration overhead compounds. Every new team, business unit, or data domain requires a new space with its own business logic defined from scratch. |
The maintenance burden grows. Definitions drift. Governance becomes difficult to enforce. |
What Kobai adds: a shared semantic foundation for Genie
Kobai provides the semantic layer that Genie draws from. Rather than each Genie space defining its own business logic, the enterprise’s shared concepts (entities, relationships, terminology, and rules) are defined once in Kobai’s semantic model and made available to every Genie space built on top.
The integration is direct. With a single line of code in the Kobai SDK, a Genie space is connected to the semantic model. Genie then queries semantic views — governed projections of the knowledge graph — rather than raw Delta tables. The answers Genie produces are grounded in the shared definition of what each entity means and how it relates to others.
|
// Connect a Genie space to the Kobai semantic model kobai.genie.create_space( ontology_domain = "operations", genie_space_name = "Operations Intelligence" ) |
That single call creates a Genie space backed by the full semantic context of the operations domain — entities, relationships, query views, and ontology metadata — all governed by Unity Catalog, all executing on Databricks compute.
|
Ontology Domain Rooms Kobai organizes the semantic model into Ontology Domain Rooms — domain-specific sections of the knowledge graph that correspond to distinct areas of the business (operations, finance, supply chain, asset management). Each Genie space connects to one or more Domain Rooms, inheriting the semantic context of those domains. When a question spans domains, Genie draws from the connected context across multiple rooms, enabling cross-domain reasoning without the user or the space being aware of the underlying model complexity. |
What changes when Genie has a shared semantic model
Genie answers become consistent across teams
When “revenue,” “customer,” and “active asset” are defined once in the semantic model, every Genie space that draws from that model uses the same definitions. A finance team and a sales team asking the same question get the same answer because they are querying the same semantic ground truth, not independently configured space logic.
Cross-domain questions become answerable
Because the semantic model connects entities across domains — assets to maintenance events to engineers to operational schedules, or customers to contracts to products to support history — Genie can traverse those connections to answer questions that a single-domain space cannot. The question “what is the projected revenue impact if the turbine maintenance on Site B extends beyond the planned window?” requires connecting assets, maintenance, operational schedule, and revenue data. With a connected semantic model, Genie can answer it.
Genie scales without rebuilding context
Each new team or business unit that needs a Genie space connects to the appropriate Domain Room of the shared semantic model, rather than defining business logic from scratch. The configuration overhead that would otherwise compound with every new space is reduced to a declaration of which domain context is relevant. The semantic model is maintained centrally, by the domain experts who own each area, and evolves as the business evolves.
AI answers carry explainable lineage
Genie answers grounded in a Kobai semantic model are traceable. Every answer can be resolved back through the semantic query to the specific entities, relationships, and data records that produced it. For teams in regulated environments or simply for teams that want to trust the answers they are acting on, this traceability is the difference between a tool they use cautiously and one they rely on.
|
Genie gives your teams natural language access to Lakehouse data. Kobai gives Genie the shared business context needed to make that access accurate, consistent, and explainable at enterprise scale. |
What this looks like in practice
|
Energy / Industrials: Cross-domain operational intelligence An operations leader asks Genie: “Which of our upstream assets are at highest failure risk this month, and what is the projected production impact if they go unscheduled?” Without a shared semantic model, this question crosses the boundary between the asset management space, the maintenance space, and the production forecasting space, and no single Genie space can answer it. With Kobai’s semantic model connecting assets, maintenance history, failure modes, and production schedules, the question is answered in a single conversation: risk-ranked assets, associated maintenance windows, and projected production exposure in the language of the operations domain, not the schema. |
|
Professional services: Consistent commercial intelligence across business units A commercial director wants every account manager in the business to be able to ask Genie questions about their client relationships: contract status, renewal risk, cross-sell opportunities, engagement history. Without a shared semantic model, each regional team builds its own Genie space with its own definition of “active client,” “reported revenue,” and “engagement.” Answers are inconsistent. Global reporting cannot reconcile them. With a shared Kobai semantic model, every regional Genie space draws from the same definition of every commercial entity. Global and regional answers are consistent. The commercial director can trust the cross-regional picture. |
|
Manufacturing: Supply chain to production intelligence A supply chain manager asks: “Which supplier delays this week will affect which production lines, and by how much?” This question requires traversing supplier records, inbound logistics, parts inventory, BOM relationships, and production schedules across systems that were designed independently. With a Kobai semantic model connecting all of these entities, Genie follows the relationship chain from delayed supplier to affected parts to impacted production lines to quantified schedule risk. The answer that previously required a data analyst and half a day resolves in the conversation. |
How the Kobai + Genie architecture works
The Kobai + Genie architecture is additive. Genie remains unchanged. Kobai adds the semantic layer on top of the Databricks Lakehouse, and Genie connects to it through semantic views published by the Kobai SDK.
|
Layer |
What it does |
|
Databricks Lakehouse(Delta · Unity Catalog · Compute) |
Governed data foundation. All Kobai semantic queries execute on Databricks compute. Unity Catalog access controls govern which data each user’s Genie conversation can access. There is no data movement or replication. |
|
Kobai Studio(Semantic model authoring) |
Domain experts define the shared semantic model: enterprise entities, relationship types, and business rules. No code required. Changes take effect immediately across all connected Genie spaces. |
|
Kobai Saturn(Semantic graph layer) |
Builds the knowledge graph index directly over governed Delta tables. Publishes semantic views and query views that Genie spaces connect to. Graph traversals execute as SQL on Databricks compute. |
|
Ontology Domain Rooms |
Domain-specific sections of the shared semantic model (operations, finance, supply chain, etc.). Each Genie space connects to one or more rooms, inheriting governed context without redefining it. |
|
Genie Spaces(Kobai SDK — one line of code) |
Genie spaces connect to Kobai semantic context via a single SDK call. Genie queries semantic views rather than raw tables. Answers are grounded in the shared semantic model, governed by Unity Catalog, and explainable via Episteme lineage. |
The key architectural property is that the semantic model is the shared asset — defined once, maintained by domain experts, and consumed by every Genie space, AI agent, or analytics tool that connects to it. When the model is updated, every connected Genie space reflects the change. There is no per-space business logic to keep in sync.
Genie without Kobai vs Genie with Kobai
|
Capability |
Genie (standalone) |
Genie + Kobai semantic layer |
|
Business term consistency |
Defined per space. Drift occurs as spaces multiply. |
Defined once in the semantic model. Consistent across all spaces and teams. |
|
Cross-domain questions |
Limited to the context of a single space. |
Traverses the semantic model across domains in a single conversation. |
|
Scaling to new teams |
Requires rebuilding business logic per space. |
New spaces connect to existing Domain Rooms. Configuration is additive. |
|
Answer explainability |
Genie cites data sources; lineage to entity-level reasoning varies. |
Full graphical lineage from answer to semantic query to source entity via Episteme. |
|
Governance |
Unity Catalog governs raw table access. Space-level filtering may vary. |
Unity Catalog governs at semantic query execution. Consistent across all spaces. |
|
Maintenance over time |
Business logic must be updated in every space as definitions evolve. |
Update the semantic model once. All connected spaces inherit the change. |
GETTING STARTED
Starting with the Genie Spaces Accelerator Kit
Kobai’s Genie Spaces Accelerator Kit is available on the Databricks Marketplace and provides a structured path to deploying a shared semantic foundation for Genie within an existing Databricks environment.
- Define a shared semantic model for a focused starting domain — typically a domain where multiple teams are using Genie and definition drift is already causing problems
- Connect 1–2 Genie spaces to the semantic model using the Kobai SDK
- Demonstrate cross-domain question capability on a live dataset before expanding to additional domains
- Expand incrementally: each new Domain Room connected makes every existing space richer through the network effects of shared entities
The typical path from environment setup to a working multi-domain Genie space grounded in a shared semantic model is measured in weeks, not months. The Kobai team supports the semantic modelling process and the Genie space connection — domain experts author the ontology, engineers handle the data connections.
Genie at enterprise scale starts with shared context
The organizations that get the most from Genie are not the ones with the most spaces. They are the ones where every Genie space draws from a shared, governed, semantically rich understanding of the enterprise — one that reflects how the business actually thinks about its entities, relationships, and rules.
Kobai provides that foundation on the Databricks Lakehouse. The semantic model is built by domain experts, governed by Unity Catalog, and made available to Genie through a single SDK integration. Cross-domain questions become answerable. Definitions stay consistent. AI answers carry explainable lineage. And every new Genie space added to the enterprise makes the shared model more valuable, not more expensive to maintain.
To explore how a shared semantic model accelerates Genie at enterprise scale, visit kobai.io/databricks or contact us at contact@kobai.io. The Genie Spaces Accelerator Kit is available now on the Databricks Marketplace.
COMMENTS
