From requirement to production

Describe the outcome. Build the agent.

OniLeo provides a structured path from a business requirement to a tested, published and continuously improved enterprise AI agent. Start with what the business needs the agent to accomplish, not with prompts, models or a blank workflow canvas.

01

DESCRIBE

02

BLUEPRINT

03

GENERATE

04

REFINE

05

CONNECT

06

TEST

07

PUBLISH

08

IMPROVE

01 · Describe

Start with the customer outcome.

Describe the business requirement in natural business language. Focus on what customers need to accomplish, what information the agent may need, what questions it should answer, and when business actions or human assistance may be required.

Create a clinic assistant that can answer clinic questions, book or reschedule appointments, cancel bookings and transfer customers when human assistance is required.

Start in business language, not prompts, nodes or model configuration.

Business Goal

Customer Journeys

Operational Requirements

OniLeo Analysis

02 · Blueprint

Turn the requirement into a structured Agent Blueprint.

OniLeo translates the requirement into a structured design for the agent. The Blueprint identifies what the agent needs to understand, what customer journeys it must support, what information and knowledge are required, and where decisions, actions or handoffs are needed.

Capabilities

The business outcomes the agent must support.

Customer Intents

The customer needs and requests associated with those capabilities.

Information Requirements

The information required to complete structured customer journeys.

Knowledge Requirements

The approved business information the agent needs to answer relevant questions.

Workflow Requirements

The business journeys required for each capability.

Decision Requirements

Points where the journey must branch based on customer or business conditions.

API / Action Requirements

External operations required to complete a process.

Handoff Requirements

Situations where human assistance is required.

Example blueprint capabilities

Book AppointmentReschedule AppointmentCancel AppointmentClinic InformationHuman Assistance
Unstructured Business RequirementStructured Agent Blueprint

03 · Generate

Generate executable workflows from the Blueprint.

OniLeo uses the Agent Blueprint to generate the workflows required for each capability. Generated workflows can include information capture, decision points, enterprise knowledge steps, business actions, messages and human handoffs.

THE BLUEPRINT GENERATES THE WORKFLOWS.

Example 1 · Book Appointment

  • Start
  • Capture NameInformation
  • Preferred DoctorInformation
  • Preferred DateInformation
  • Check AvailabilityAPI Action
  • Available?Decision
  • Yes · Confirm AppointmentMessage
  • No · Offer AlternativeDecision path

Example 2 · Clinic Information

Customer QuestionEnterprise KnowledgeAnswerContinue Appropriate Journey

04 · Refine

Keep enterprise teams in control of the generated agent.

Generation accelerates the starting point and enterprise review provides control. Teams can inspect the generated workflows and refine information requirements, decision paths, messages, knowledge mappings, actions and handoffs before the agent is approved for testing.

Generated AutomaticallyEnterprise ReviewApproved Configuration
  • Change a customer question
  • Mark required information
  • Adjust a decision branch
  • Map a knowledge requirement
  • Configure an API action
  • Define a handoff path
Explore Workflow Studio →

05 · Connect

Connect the agent to the knowledge and systems it needs.

Add the approved enterprise information identified by the Blueprint and connect configured business actions required by the customer journeys. This gives the agent both the context to answer and the ability to participate in structured business processes.

Enterprise Knowledge

  • Documents
  • Websites
  • Approved Business Content

OniLeo Agent

Business Actions

  • Configured APIs
  • Business Systems
  • Approved Operations

Active journey: booking an appointment. The customer asks “What time does the clinic close?” The agent answers from approved enterprise knowledge, then continues the appointment journey.

Knowledge can support the conversation without discarding the active customer journey.

06 · Test

Test the conversations customers will actually have.

Before publishing, validate how the agent responds to expected journeys, variations, interruptions, missing information, knowledge questions and supported language changes.

Scenario A

I want to book an appointment tomorrow.

Recognise the appointment request and start the appropriate workflow.

Scenario B

I do not have my patient ID.

Handle unavailable information according to the configured workflow.

Scenario C

Before that, what time does the clinic close?

Answer from approved enterprise knowledge without losing the active journey.

Scenario D

Can we continue in Hindi?

Adapt the conversation language while preserving relevant context.

Pass / Review / FailRefine If RequiredRetest
Explore Test Studio →

07 · Publish

Move the validated agent into production.

Once the agent version has been reviewed and validated, publish the approved version to the required customer channels while keeping work-in-progress configuration separate from the live experience.

Validated Agent VersionPublishWeb ChatBrowser VoiceSIP / Phone Voice

One approved agent configuration across the channels selected for the business use case. Working Version is never the same as Published Version.

08 · Improve

Use production interactions to improve the next release.

After publishing, operational signals help teams identify weak scenarios, knowledge gaps, workflow issues, handoff patterns and runtime behaviour that may require refinement. Changes can then be reviewed, retested and published as the next agent version.

OperateObserveImproveTestPublish