Multi-Agent LLM System for Business Process Automation

cognizant · Tech · 2024 · source Agentic: yes · promotional bias: med · summary-only (no fetched source) · extracted with prompt 0.5.6 (schema v2)

Generated from extraction.json — do not edit by hand. Every claim cites verbatim source evidence; stated means the quote directly asserts the claim, implied means it supports it indirectly.

Problem

Domain: Enterprise business process automation

Cognizant wanted business users to create and deploy AI-powered decision workflows without requiring deep data-science or engineering expertise.

Constraints:

Business stakeholders needed to influence workflows aligned with their KPIs.
The platform needed to run in cloud, client environments, and air-gapped deployments.
Agents needed to call APIs, run commands, execute code, and collaborate.
Explainability and human oversight had to remain part of the workflow.

Architecture

Neuro AI is a multi-agent LLM platform in which agents combine an LLM brain with tools and capabilities, communicate with each other, and support workflows such as intranet search, use-case scoping, synthetic-data generation, and automated process creation. It can be hosted by Cognizant or deployed on premises, including air-gapped environments with commercial or open-source models.

Diagram confidence: reconstructed (reconstructed from prose).

flowchart LR
  business-user-interface["Business User Interface"]
  agent-collaboration-layer["Agent Collaboration Layer"]
  use-case-discovery-agents["Use Case Discovery Agents"]
  workflow-creation-agents["Workflow Creation Agents"]
  department-agents["Department Agents"]
  business-user-interface --> agent-collaboration-layer
  agent-collaboration-layer --> use-case-discovery-agents
  agent-collaboration-layer --> workflow-creation-agents
  agent-collaboration-layer --> department-agents
  use-case-discovery-agents --> agent-collaboration-layer
  workflow-creation-agents --> agent-collaboration-layer
  department-agents --> agent-collaboration-layer

Drag to pan · scroll or use +/− to zoom · reconstructed from the source's described component topology

Components

Component	Role	Interacts with
`Business User Interface`	Lets business users create and operate AI workflows.	`Agent Collaboration Layer`
`Agent Collaboration Layer`	Coordinates communication among agents for complex processes.	`Use Case Discovery Agents`, `Workflow Creation Agents`, `Department Agents`
`Use Case Discovery Agents`	Help identify and scope AI use cases.	`Agent Collaboration Layer`
`Workflow Creation Agents`	Create automated decision and process workflows.	`Agent Collaboration Layer`
`Department Agents`	Handle domain-specific search or process tasks such as HR, IT, or Legal.	`Agent Collaboration Layer`

Data flow

A business user identifies or scopes a workflow through Neuro AI. Specialized agents can discover use cases, generate synthetic data, create workflows, and delegate tasks to domain agents. In intranet search, a main search agent delegates to department agents such as HR, IT, and Legal, which can further delegate to more specialized agents.

“in their intranet search implementation, a main search agent can delegate queries to specialized department agents (HR, IT, Legal, etc.)”summary, stated

Why this architecture

Cognizant designed Neuro AI as a multi-agent platform because enterprise process automation requires agents that can act through tools, delegate to specialist domains, and run in varied deployment environments rather than merely produce text.

“Unlike simple LLM models that just generate output, these agents can execute commands, call APIs, run code in containers, and interact with other systems.”summary, stated

Models used: commercial LLMs · open-source on-premises models

Alternatives considered (roads not taken)

Require business users to rely on data-science teams for AI workflows

Rejected because: The platform's goal was to bridge the gap between AI capability and business-user needs.

“bridging the gap between technical AI capabilities and business users' needs.”summary, stated

Offer only hosted cloud deployment

Rejected because: Clients may need on-premises or air-gapped deployments for sensitive data.

“installed within a client's environment, including air-gapped environments for sensitive data.”summary, stated

Patterns

`multi-agent-orchestration` (architecture)

Neuro AI lets agents communicate and delegate tasks across business processes.

“The platform allows agents to communicate with each other to handle complex business processes”summary, stated

`tool-use-orchestration` (execution)

Agents can call APIs, execute commands, run code, and interact with systems.

“these agents can execute commands, call APIs, run code in containers, and interact with other systems.”summary, stated

`synthetic-data-generation` (data)

The platform includes synthetic-data generation to help business users bootstrap AI workflows.

“The system includes features for opportunity identification, use case scoping, synthetic data generation, and automated workflow creation”summary, stated

Tradeoffs

Support multiple deployment modes

Chose Hosted, on-premises, and air-gapped deployments over One managed-cloud-only architecture.

Rationale: Sensitive enterprise data and client constraints require deployment flexibility.

“The platform can be deployed either as a hosted solution managed by Cognizant or installed within a client's environment”summary, stated

Use action-capable agents

Chose Agents with tools, APIs, commands, and code execution over LLMs that only generate text outputs.

Rationale: Business process automation requires acting inside systems, not only answering questions.

“Unlike simple LLM models that just generate output, these agents can execute commands, call APIs, run code in containers”summary, stated

Gotchas (what bit them)

Business users could not directly shape AI systems with traditional implementations.

Root cause: Traditional AI work required significant data science expertise.

Resolution: Neuro AI provides workflow-creation and use-case-scoping agents for business users.

“Traditional AI implementations often required significant data science expertise, making it difficult for business stakeholders to directly influence and control AI systems”summary, stated

Enterprise deployments need flexible data-boundary options.

Root cause: Some clients cannot use hosted commercial models because of sensitive data constraints.

Resolution: Neuro AI supports hosted, client-environment, air-gapped, commercial-model, and open-source-model deployments.

“This includes options for using either commercial LLMs or open-source models running completely on-premises.”summary, stated

Evaluation

The summary emphasizes explainability and human oversight but provides no quantitative public evaluation metrics.

“all while maintaining explainability and human oversight.”summary, stated

Outcomes

Cognizant created Neuro AI for business-user-driven AI workflow automation.
The platform supports intranet search, process automation, use-case scoping, and synthetic-data generation.
Deployment can run in hosted, on-premises, or air-gapped modes.

The summary does not disclose production volume or benchmark metrics.

“Cognizant developed Neuro AI, a multi-agent LLM-based system that enables business users to create and deploy AI-powered decision-making workflows”summary, stated