AI Engagement Platform

OnDimi is an experimental platform exploring how agentic, multi-agent AI can plan, orchestrate, and execute complete customer journeys across multiple channels. It serves as a personal testbed for designing and coordinating Decisioning, Planning, Orchestration, and Specialist AI Agents that work together in real time to deliver dynamic, compliant, and personalized engagement.

Note: OnDimi is a personal experimental project and is not affiliated with any company or organization.

Use Cases Explored

Hi Mary, This is Jack with onDimi Auto Insurance. You insurance expires on March, 25 2025. I am here to help you renew it. ready?

AI Insurance Renewal Agent

Assist customers through a smooth renewal experience by confirming their readiness, answering questions, and offering flexible next-step options.

Hi Angie, Your washing machine installation is scheduled for Friday, May 10 at 10 AM. You can confirm, reschedule, or cancel your appointment here.

Ai Customer Service

Engage with customers to ensure a seamless setup experience, confirming their availability and providing rescheduling options.

Hi John, It’s onDimi Bank. Your loan quote is has been pre-approved. Would you like to go over the details with a Lending Specialist?

AI Lending Specialist

Proactively reach out to loan applicants about their online pre-approval, address any questions, and schedule a meeting with a lending specialist.

Planning AI Agent - Generated Plan - Timeline view

DYNAMIC DECISIONING AND PLANNING

Decisioning and Planning AI

The Decisioning and Planning AI is a decision-driven, agentic intelligence that not only schedules customer interactions but also determines when, how, and which agent should engage the customer. It continuously refines the communication plan based on customer data, behavior patterns, compliance rules, business objectives, and journey context. The agent generates personalized, multi-day, multi-channel action plans—spanning channels, languages, and specialized AI agents—and adapts them in real time as conditions change.

Personalized Action Plans: Generates multi-day, multi-channel customer engagement plans tailored to individual needs, behaviors, and history—setting up future conversations and next best actions.
Intelligent Decisioning: Determines the optimal timing, channel, content, and which AI or human agent should engage the customer at every step.
Dynamic Adaptation: Continuously updates plans in real time based on customer interactions, business rules, compliance constraints, and evolving journey context.
Agentic Orchestration Guidance:: Provides the Orchestrator AI with structured, step-by-step plan guidance, enabling smooth collaboration between specialist agents across SMS, email, and voice.

Multi-Agent

Multi-Agent Architecture & Orchestration

OnDimi is built on a multi-agent, agentic AI architecture where multiple specialized agents work together—each with its own role, intelligence, and capabilities. Instead of a single model handling everything, the system coordinates a network of agents that collaborate, hand off tasks, and execute complex customer journeys end-to-end.

multi-agent builder - new multi-node - existing ai agents

Collaborating AI Agents: Multiple agents: planning, conversational, and specialist—work together to drive each customer journey.
Role-Based Intelligence: Each agent handles its own responsibilities, from decision-making to customer interactions to domain expertise.
Dynamic Handoffs: Agents seamlessly pass tasks or conversations to the right agent or sub-agent at the right moment.
Omnichannel Orchestration: Supports coordinated engagement across SMS, email, voice, and system actions.

AI Agent Builder

Build intelligent agents visually with a flexible, drag-and-drop builder. The AI Agent Builder lets you assemble powerful agents by connecting the core building blocks they need—data, knowledge, guardrails, tools, and MCP integrations. You can reuse existing resources or create new ones, making it easy to design agents that are consistent, capable, and tailored to their role within the multi-agent architecture.

Visual Agent Builder: Design AI agents with a drag-and-drop canvas, connecting nodes into multi-step flows in seconds.
Reusable Building Blocks: Attach or reuse existing data, knowledge, guardrails, tools, and MCPs—or create new ones when needed.
Structured Capabilities: Give each agent exactly the right context and capabilities by combining multiple building blocks in one place.
Smart Node Management: Clearly see which blocks are already connected (grayed out), making it easy to extend agents without duplicating work.

AI Agent Building Blocks

Every AI agent in OnDimi is powered by a set of modular, configurable building blocks. These components define what the agent knows, what rules it must follow, what actions it can take, and what external systems it can access. By combining and reusing these blocks, you can design agents that are precise, safe, and deeply aligned with your business workflows.

Data Sources

Give each AI agent access only to the data it needs—nothing more.
Agents can reference structured fields such as customer data, business data, appointment details, demographic attributes, and use-case-specific fields.
You choose exactly which fields are available, ensuring privacy, relevance, and contextual accuracy during customer interactions.

Guardrails

Define the rules an agent must follow.

Guardrails can enforce:

Input validation (e.g., detect jailbreak attempts or unsafe prompts)
Output constraints (e.g., preventing hallucination or enforcing factual responses)
Business guardrails such as business policies, compliance rules, or regulatory boundaries

This ensures every agent operates safely, ethically, and within your organization’s requirements.

Knowledge Bases

Provide contextual knowledge that the agent can use to answer questions.
OnDimi supports two types:

Persistent Knowledge: Always available to the agent (e.g., product details, business rules, policies)
Retrieval-Based Knowledge: Fetched only when relevant, using Retrieval-Augmented Generation (RAG) for fresh, targeted responses
This lets agents combine long-term domain expertise with dynamic, context-specific information.

Tools (Function Calls)

Extend an agent’s capabilities with custom tools.

Each tool includes:

Name and description
Input schema (parameters, required fields, data types)
Strict modes
Tool result handling, storing outputs for downstream steps
Full JSON preview of the tool definition
Tools allow agents to take real actions: scheduling callbacks, sending emails, updating records, triggering workflows, and more.

MCP (Model Context Protocol) Integrations

Connect agents to external systems through Model Context Protocol (MCP) servers.

Agents can:

Fetch available tools from MCP servers
Select only approved tools
Use multiple transport types (SSE, streamable HTTP, etc.)
Configure auth keys and access controls
Choose which tools require approval and which do not

MCP unlocks powerful integrations like Google Calendar, CRMs, ticketing systems, and internal services—directly from the agent.

AI Agents

AI agents themselves can be exposed as callable tools—allowing one agent to delegate work to another. In this pattern, a specialized agent is wrapped as a function that other agents can invoke when needed. This creates a hierarchical, team-like structure where:

An Orchestrator Agent decides when to call a specialist

Specialist Agents (e.g., Lending, Customer Service, Voice AI) execute domain-specific tasks
Each agent brings its own data, knowledge, guardrails, and tools

This mirrors real organizational collaboration: a manager coordinating experts rather than a single agent attempting everything. By treating agents as tools, the system supports modular delegation, specialization, and scalable multi-agent workflows.

Every Interaction. Every Agent. Fully Traceable.

Unified Customer Journey Visibility

Understand the complete story behind every customer journey—past, present, and future. Ondimi provides end-to-end visibility across all AI and human agent actions, customer inputs, scheduled communications, executed steps, and upcoming tasks. This transparency empowers teams to deliver faster resolutions, consistent service, and confident, informed customer conversations.

Ondimi gives your team full visibility and traceability across the entire journey—what has happened, what’s happening now, and what’s planned next—ensuring clarity, continuity, and seamless customer support.

Key Capabilities

Complete Interaction Trace
Track every customer message, AI agent reply, human action, tool execution, and data update in one unified activity log.
Multi-Agent Transparency
See which AI agents were involved, when they acted, why they acted, and what impact their decisions had.
Real-Time Journey Monitoring
View live statuses of ongoing conversations, scheduled actions, and pending steps across channels.
Timeline View
Explore a chronological, human-readable narrative of all communications—SMS, email, voice, and system actions.
Calendar View
Visualize the entire communication plan—executed, upcoming, cleared, or rescheduled—at a glance.
Action Outcomes & State Changes
Instantly see whether actions were executed, skipped, canceled, or cleared, and how customer data was updated.
Human + AI Collaboration Insight
Understand exactly where humans stepped in, where AI operated autonomously, and how both contributed to the journey.

INBOUND & OUTBOUND CALL

Enhancing Voice Call Handling

Experiementing the use of GenAI, Whisper model for speech recognition, Google TTS, and openAI LLM to enhance the conversational experience in voice calls. Watch the Video

Personalization: Personalizes interactions by recognizing callers
Call Topic: Detecting the call topic to be discussed in a voice call.
Answering queries: Easily and accurately answer queries over the phone without the need fot an agent involvement.
Transcription: Transcribing the call on the fly while the call is going on.
Scheduling call/appointment: Scheduling, Rescheduling or canceling a call or appointment. Transfering to agent if required.
Setting follow up actions: Setting reminders or a new plan based on the outcome of the call.