Does DataGOL evaluate agents before changing models?

This is an important part of the architecture being developed. Because changing an LLM can change agent behavior, DataGOL is developing an evaluation framework for testing agents before deploying a model change.

The goal is to evaluate things such as:

• Task completion

• Tool selection

• Reasoning/workflow behavior

• Output quality

• Groundedness

• Regression against existing use cases

This allows model changes to be treated more like a controlled software release than simply changing an API configuration.

Does DataGOL support an ontology or knowledge graph?

Yes. DataGOL can represent relationships between entities and information sources through its semantic and knowledge-graph capabilities.

This is particularly useful when answering questions that require navigating relationships across multiple systems.

For example: Customer → Account → Orders → Products → Support Tickets → Contracts

If we start with Snowflake, can we later move to Databricks?

Yes, DataGOL is designed to minimize infrastructure lock-in.

The DataGOL semantic, agent and intelligence layers can operate above the underlying data systems.

Where data remains in the existing warehouse, DataGOL can use the appropriate connector and push queries to that system.

This means changing the underlying data platform does not inherently require rebuilding the entire agent architecture.

Can DataGOL connect to multiple data platforms simultaneously?

Yes. DataGOL supports multiple connectors and can work across heterogeneous data environments.

An enterprise can, for example, have:

Snowflake + Databricks + Salesforce + SharePoint + REST APIs

within the same overall DataGOL environment.

DataGOL is designed to provide a single intelligence layer across those environments, rather than forcing the organization to consolidate everything into one platform first.

Can DataGOL work with existing MDM systems?

Yes, subject to the customer's architecture.

DataGOL does not require an enterprise to replace its existing MDM strategy.

Existing MDM systems can remain authoritative for entities such as:

• Customer

• Product

• Supplier

• Employee

• Account

DataGOL can then use that governed information as part of the semantic/context layer used by agents.

The exact entity-resolution and survivorship rules remain dependent on the customer's MDM implementation.

Can I swap the underlying model?

Yes. Model choice is configurable.

Different models can be selected for different agents or workloads. For example, one use case may prioritize reasoning quality, another latency or cost, and another may require an open-weight or privately deployed model.

In short: models are swappable, but enterprise-grade model portability requires evaluation rather than simply changing an endpoint.

Can non-technical users utilize the platform effectively?

Yes, our platform is designed for both technical and non-technical users, featuring intuitive interfaces, conversational analytics, and automation that make it accessible for anyone to derive insights and make data-driven decisions.

Does DataGOL fine-tune foundation models?

Not currently as a standard part of the platform.

DataGOL's current approach focuses on context engineering, semantic modeling, retrieval, orchestration, tools and governance, rather than requiring model fine-tuning for each customer.

This has an important architectural advantage: organizations can change models without having to retrain a proprietary model every time their underlying model strategy changes. Fine-tuning may be appropriate for specific future or custom use cases, but it is not required for the core DataGOL architecture.

Does DataGOL have a semantic layer?

Yes. DataGOL's architecture includes a semantic/context layer that provides agents with business meaning rather than requiring them to reason directly over raw tables and documents.

The semantic layer can provide context around:

• Business entities

• Relationships

• Metrics

• Business definitions

• Data sources

• Metadata

• Relationships between structured and unstructured information

This allows an agent to reason about concepts such as Customer, Product, Order, Revenue or Account rather than simply seeing disconnected database columns.

Does DataGOL lock customers into a particular data warehouse?

No. DataGOL is designed to sit above the existing data infrastructure.

DataGOL can work with existing warehouses, databases, APIs, files, and unstructured-data repositories.

Where appropriate, DataGOL can push queries down to the underlying database or warehouse, allowing organizations to continue using their existing data investments. The goal is to provide a common intelligence and agent layer without requiring an enterprise to replace its existing data platform.

Does DataGOL replace our existing data warehouse?

Not necessarily. DataGOL can sit above existing warehouses and databases and leverage the investments an organization has already made.

It can also participate in data ingestion, transformation and orchestration when that is appropriate. Therefore DataGOL should not be viewed simply as another warehouse or lakehouse. Its role is to provide an intelligence, semantic and agent layer across the enterprise data estate.

Does DataGOL require us to move all our data into DataGOL?

No. DataGOL can operate against existing data platforms using connectors and query pushdown.

It can also ingest and orchestrate data into open storage formats such as Iceberg/Parquet on S3 or HDFS, depending on the architecture and use case.

This gives organizations a choice between:

Query in place and Ingest, transform and manage data through the DataGOL data pipeline architecture.

Does DataGOL support data versioning and lineage?

Yes. DataGOL's data architecture supports versioned data and lineage, including architectures based on open table/storage formats such as Iceberg and Parquet.

This enables organizations to understand:

• Where data came from

• How it was transformed

• Which version is being queried

• How changes propagate through the data pipeline

This is particularly important for AI because an agent should not unknowingly combine information from incompatible versions of an entity or dataset.

Does DataGOL support RAG?

Yes. DataGOL supports retrieval-augmented generation (RAG) for enterprise knowledge.

DataGOL can ingest and index unstructured content and make it available to agents through retrieval workflows.

Supported content can include sources such as:

• Documents

• PDFs

• Confluence

• SharePoint

• Google Drive

• Files

• Audio

• Video

• Other enterprise knowledge sources

  
  

Does DataGOL use only RAG?

No. RAG is one component of the overall architecture. For structured enterprise data, DataGOL can query the underlying systems directly. For unstructured information, it can use retrieval and RAG workflows.

This allows an agent to combine: Structured data + semantic context + unstructured knowledge + business relationships within a single workflow.

Does DataGOL validate whether an answer is grounded?

Yes. DataGOL can use a separate model-based validation step to assess the groundedness of the response against the information retrieved or generated during the workflow.

This creates a generation → validation pattern rather than assuming that the first model response is automatically correct.

For high-value enterprise applications, this can be combined with additional deterministic checks, business rules and evaluation datasets.

How does DataGOL determine which customer or product record is authoritative?

DataGOL is designed to preserve source and lineage context, but authoritative-record governance is implementation-dependent.

For enterprise deployments, it is important to distinguish between:

• System of record

• System of reference

• Master data

• Derived/analytical data

• Historical versions

DataGOL can use metadata, lineage, semantic definitions and source-system context to understand where information originated and how it was transformed.

Where an enterprise has an established MDM or authoritative source, that source can be incorporated into the data architecture and semantic model.

How does DataGOL reduce hallucinations?

DataGOL uses multiple layers of grounding rather than relying solely on the foundation model.

These can include:

• Retrieval from authoritative enterprise sources

• Semantic/business context

• Structured-data queries

• Source citations

• Tool-based execution

• Agent policies and guardrails

• Validation of generated responses

• Groundedness checks

The objective is to ensure that the model is not simply generating a plausible answer, but is producing an answer grounded in enterprise data and context.

How does DataGOL support security and compliance?

Data stays within your infrastructure. DataGOL respects all existing access rules (e.g., Snowflake row-level permissions) and provides enterprise-grade controls like SSO, audit logs, data masking, and full role-based access.

How does the platform ensure data security and compliance?

We prioritize data security and compliance by implementing stringent data management practices, encryption, and access controls, ensuring your data is protected according to the latest industry standards and regulations.

How is DataGOL priced?

DataGOL offers flat, predictable licensing—no surprise compute bills, no per-query charges. This is ideal for both cost-sensitive SaaS startups and budget-conscious enterprise teams.

Is DataGOL tied to a specific LLM or foundation model?

No. DataGOL is model-agnostic. DataGOL is designed as an orchestration and intelligence layer that can work with different foundation models depending on the use case, enterprise requirements, performance, cost, and deployment constraints.

This includes commercial models as well as open-weight models where required. Model selection can be configured at the agent/use-case level rather than requiring the entire DataGOL architecture to be built around a single model.

The result: enterprises can evolve their model strategy without having to rebuild their data and semantic architecture.

What happens if our data architecture changes in the future?

DataGOL is designed around portability at multiple layers:

Model portability - Change foundation models without redesigning the entire application.

Data-platform portability - Connect to different warehouses and databases.

Knowledge portability - Connect structured and unstructured enterprise sources.

Semantic portability - Keep business concepts and relationships above individual physical data implementations.

Agent portability - Evaluate agent behavior when underlying models or data environments change.

This reduces the risk that an enterprise's AI strategy becomes permanently coupled to a particular model vendor or data platform.

What support and training resources are available for new users?

We offer comprehensive support and training resources, including detailed documentation, online tutorials, and dedicated customer support teams to ensure you can maximize the value of our platform for your business needs.