GenAI Data Fusion Overview
K2view GenAI Data Fusion (AI Fusion) is an end-to-end platform for building, testing, running and governing enterprise AI agents on top of K2view Data Products. Agents built with AI Fusion answer from real-time, entity-level business data — a single customer, account, policy or member — retrieved securely from across your enterprise systems.
AI Fusion is purpose-built for high-scale 360° systems in industries such as telecommunications, finance, insurance and healthcare, where privacy, performance and precision are essential.
This documentation describes AI Fusion 3. If you are running AI Fusion 1.x or 2.x, see the legacy documentation.
The Challenges Enterprises Face in Adopting AI
Foundation models are trained on extensive general knowledge, but they have no access to your organization's specific, real-time operational data — the data needed for meaningful business interactions.
Industry research shows that 50–70% of customer service inquiries require direct access to enterprise systems. Questions such as "What's my current account balance?" or "Why was my bill higher this month?" depend on real-time data that is spread across several business systems.
Meeting this need raises several data challenges:
- Data Fragmentation — enterprise data resides across dozens of disparate systems (CRM, Billing, ERP, support platforms), making it difficult to deliver complete, unified information to AI applications.
- Real-time Requirements — AI applications need quick access to up-to-date data to deliver accurate, responsive conversations.
- Security & Privacy — LLMs must access only the data relevant to the current user and context, under strict access control.
- Scale & Performance — thousands of simultaneous AI interactions must be served with fast responses across enterprise-scale data volumes.
- Quality & Reliability — responses must rely on complete, current and accurate data to prevent hallucinations and preserve user trust.
How AI Fusion 3 Addresses Them
AI Fusion combines the Data Product Platform with a declarative agent runtime, a dedicated development environment in Studio, and built-in evaluation and observation.
Declarative Agents on Data Products
In AI Fusion 3, an agent is a set of files that lives inside a Data Product (LU): an agent.yaml definition, plus optional skills, sub-agents, reference documents, evaluation suites and observation configuration. There is no orchestration flow to build — the runtime runs the reasoning loop, calls tools, delegates to sub-agents, streams the answer, and records everything for trace and analysis.
Agents work directly on the entity's micro-database: a step_start hook loads the customer's data and the built-in query tool lets the LLM run SQL against it, guided by schema skills that describe the tables. Learn more in Agent Platform.
Quality & Accuracy
- Entity-centric data — a unified 360° view of each business entity (customer, order, product), consolidated from multiple source systems.
- Live data synchronization — current information through Fabric's synchronization with the source systems.
- Schema-aware SQL generation — schema skills can embed the live LU table definitions, so the LLM always generates SQL against the deployed structure.
- Procedures and knowledge — plan-then-execute mode guided by written procedures, and cited answers from reference documents.
Privacy & Security
- Entity-level access — each conversation works on the micro-database of the entity it was started for.
- Trusted inputs — values such as the customer ID are passed as agent inputs that the LLM cannot change.
- Guardrail hooks —
pre_tool_callhooks can block any tool call;constraintsdeclare rules the agent must never break, and they cannot be dropped by delegation. - Data masking support — PII can be masked inside the tool and hook flows that read it, using Fabric masking. Masking is configured per project and is not applied automatically.
- Role-based permissions — AI Fusion web services run under standard Fabric authentication and authorization.
Control & Trace
- Full execution trace — every step, LLM call, tool call and sub-agent run, with inputs, outputs, durations and tokens.
- Cost management — token usage per LLM call, interface and agent.
- Advanced observability — LLM-scored signals per conversation, per-agent dashboards and conversation review.
Maintainability
- Unified platform — one toolset for data retrieval, agent development, testing and monitoring, inside K2view Studio.
- Form-based agent editor — build and test agents in AI Fusion Studio, with AI-assisted agent building and answer diagnosis.
- Flexible model management — Anthropic, AWS Bedrock, OpenAI, Google GenAI and Ollama connectors; choose the model and reasoning level per agent.
- Integrated evaluation — regression test suites per agent, scored by LLM-as-a-judge on configurable dimensions and thresholds.
Platform Modules
| Module | What it provides |
| Agent Platform | The declarative agent runtime: agent definitions, tools, skills, sub-agents, planning, reference search, hooks, sessions, the streaming Agent API and LLM connectors. |
| AI Fusion Studio | The Studio extension for building agents: AI Explorer, Agent Editor, Preview chat with trace, Eval Editor and AI-assisted development. |
| Evaluation | Per-agent regression test suites, LLM-as-a-judge scoring and performance baselines, run from Studio. |
| Observation | Production monitoring: automatic conversation analysis, per-agent dashboards and conversation review. |
Altogether, AI Fusion uses K2view's proven Data Product Platform to turn enterprise data into AI-ready resources, combining real-time data integration, enterprise-grade security and a governed agent lifecycle.
Next article: What's New in AI Fusion 3
GenAI Data Fusion Overview
K2view GenAI Data Fusion (AI Fusion) is an end-to-end platform for building, testing, running and governing enterprise AI agents on top of K2view Data Products. Agents built with AI Fusion answer from real-time, entity-level business data — a single customer, account, policy or member — retrieved securely from across your enterprise systems.
AI Fusion is purpose-built for high-scale 360° systems in industries such as telecommunications, finance, insurance and healthcare, where privacy, performance and precision are essential.
This documentation describes AI Fusion 3. If you are running AI Fusion 1.x or 2.x, see the legacy documentation.
The Challenges Enterprises Face in Adopting AI
Foundation models are trained on extensive general knowledge, but they have no access to your organization's specific, real-time operational data — the data needed for meaningful business interactions.
Industry research shows that 50–70% of customer service inquiries require direct access to enterprise systems. Questions such as "What's my current account balance?" or "Why was my bill higher this month?" depend on real-time data that is spread across several business systems.
Meeting this need raises several data challenges:
- Data Fragmentation — enterprise data resides across dozens of disparate systems (CRM, Billing, ERP, support platforms), making it difficult to deliver complete, unified information to AI applications.
- Real-time Requirements — AI applications need quick access to up-to-date data to deliver accurate, responsive conversations.
- Security & Privacy — LLMs must access only the data relevant to the current user and context, under strict access control.
- Scale & Performance — thousands of simultaneous AI interactions must be served with fast responses across enterprise-scale data volumes.
- Quality & Reliability — responses must rely on complete, current and accurate data to prevent hallucinations and preserve user trust.
How AI Fusion 3 Addresses Them
AI Fusion combines the Data Product Platform with a declarative agent runtime, a dedicated development environment in Studio, and built-in evaluation and observation.
Declarative Agents on Data Products
In AI Fusion 3, an agent is a set of files that lives inside a Data Product (LU): an agent.yaml definition, plus optional skills, sub-agents, reference documents, evaluation suites and observation configuration. There is no orchestration flow to build — the runtime runs the reasoning loop, calls tools, delegates to sub-agents, streams the answer, and records everything for trace and analysis.
Agents work directly on the entity's micro-database: a step_start hook loads the customer's data and the built-in query tool lets the LLM run SQL against it, guided by schema skills that describe the tables. Learn more in Agent Platform.
Quality & Accuracy
- Entity-centric data — a unified 360° view of each business entity (customer, order, product), consolidated from multiple source systems.
- Live data synchronization — current information through Fabric's synchronization with the source systems.
- Schema-aware SQL generation — schema skills can embed the live LU table definitions, so the LLM always generates SQL against the deployed structure.
- Procedures and knowledge — plan-then-execute mode guided by written procedures, and cited answers from reference documents.
Privacy & Security
- Entity-level access — each conversation works on the micro-database of the entity it was started for.
- Trusted inputs — values such as the customer ID are passed as agent inputs that the LLM cannot change.
- Guardrail hooks —
pre_tool_callhooks can block any tool call;constraintsdeclare rules the agent must never break, and they cannot be dropped by delegation. - Data masking support — PII can be masked inside the tool and hook flows that read it, using Fabric masking. Masking is configured per project and is not applied automatically.
- Role-based permissions — AI Fusion web services run under standard Fabric authentication and authorization.
Control & Trace
- Full execution trace — every step, LLM call, tool call and sub-agent run, with inputs, outputs, durations and tokens.
- Cost management — token usage per LLM call, interface and agent.
- Advanced observability — LLM-scored signals per conversation, per-agent dashboards and conversation review.
Maintainability
- Unified platform — one toolset for data retrieval, agent development, testing and monitoring, inside K2view Studio.
- Form-based agent editor — build and test agents in AI Fusion Studio, with AI-assisted agent building and answer diagnosis.
- Flexible model management — Anthropic, AWS Bedrock, OpenAI, Google GenAI and Ollama connectors; choose the model and reasoning level per agent.
- Integrated evaluation — regression test suites per agent, scored by LLM-as-a-judge on configurable dimensions and thresholds.
Platform Modules
| Module | What it provides |
| Agent Platform | The declarative agent runtime: agent definitions, tools, skills, sub-agents, planning, reference search, hooks, sessions, the streaming Agent API and LLM connectors. |
| AI Fusion Studio | The Studio extension for building agents: AI Explorer, Agent Editor, Preview chat with trace, Eval Editor and AI-assisted development. |
| Evaluation | Per-agent regression test suites, LLM-as-a-judge scoring and performance baselines, run from Studio. |
| Observation | Production monitoring: automatic conversation analysis, per-agent dashboards and conversation review. |
Altogether, AI Fusion uses K2view's proven Data Product Platform to turn enterprise data into AI-ready resources, combining real-time data integration, enterprise-grade security and a governed agent lifecycle.
Next article: What's New in AI Fusion 3
