Building a Reusable Conversational AI Platform for Distinct Business and Experiential Use Cases
The Challenge:
Most conversational AI implementations begin as isolated chatbots built for a single website, audience, or business function.
Intari was conceived as a joint venture between myself and my wife, Candice Roma, to solve a number of usecases where we saw existing AI implementation falling short. RAG-driven AI assistants would consistently hit “walls” where their knowledgebase would not have the answer a user was looking for, and have to fall back on canned apologies. Creating edgecases and conversational paths for complicated chatbots could be incredibly time consuming and cost prohibitive. Above all, AI based tools lacked personality and empathy, and as such could become difficult to use in a variety of human-centered scenarios.
A sales assistant may need to qualify leads and schedule meetings. A support agent may need to answer questions from approved documentation. A coach may need to maintain an encouraging tone and guide a user through a structured process. An entertainment character may need to preserve personality, narrative continuity, and immersion. Although all of these experiences use conversational AI, they require different:
- personalities
- knowledge sources
- business rules
- identity requirements
- interaction styles
- success measures
- privacy and governance controls
Building each experience as a separate application would create duplicated infrastructure, inconsistent behavior, and increasing maintenance demands.
The challenge was to design a reusable platform that could support many distinct AI experiences without reducing them to the same generic chatbot.
The Opportunity:
We saw an opportunity to separate the underlying conversational infrastructure from the identity, purpose, knowledge, and behavior of each AI experience.
That distinction became the foundation of Intari.
Rather than treating every assistant as a new application, Intari was designed as a platform through which organizations could configure and deploy purpose-built AI characters across different websites and business contexts.
The platform could provide a common technical foundation for:
- conversational interfaces
- knowledge grounding
- visitor identity and context
- message routing
- source tracking
- conversation history
- reporting and analytics
Each implementation could then define its own:
- character identity
- voice and personality
- business objective
- knowledge base
- conversation rules
- stage-direction behavior
- calls to action
The goal was not to build one chatbot that could do everything.
It was to build a platform capable of supporting many focused AI experiences well.
Our Approach:
We designed Intari as a modular conversational AI platform with configurable characters, workspaces, interaction modes, and business functions.
The architecture was shaped by real implementations rather than hypothetical requirements.
As new use cases emerged, we looked for the underlying capabilities they shared and incorporated those capabilities into the platform.
We organized the system around six primary areas:
- Workspace and Character Architecture-
The platform separates the organization deploying an experience from the AI character presented to the visitor.A workspace can manage one or more characters, while each character can have its own:- name and identity
- personality and tone
- instructions and behavioral boundaries
- knowledge sources
- conversation type
- stage-direction settings
- website placement
This allows the same platform to support experiences that feel fundamentally different while remaining operationally consistent behind the scenes.
- Purpose-Built Conversation Types-
Different business objectives require different conversational behavior.We designed the platform to support multiple chat types, including:- sales
- support
- coaching
- demonstration environments
- entertainment
The conversation type influences how the agent interprets visitor intent, which information it prioritizes, how it guides the interaction, and what outcome it is intended to support.
This prevents a sales assistant, coach, and fictional character from behaving like lightly renamed versions of the same system.
- Configurable Visitor Identity-
Not every experience should require the same amount of personal information.Intari supports configurable identity modes such as:- anonymous participation
- name only
- name and email
- name, email, and company
This allows each implementation to balance personalization, business needs, privacy, and user friction.
Visitor context can be incorporated into the conversation when appropriate without making identity collection mandatory for every use case.
- Knowledge and Context Grounding-
A useful AI experience must be grounded in the right information rather than relying only on a general language model.The platform was designed to support:- business-specific knowledge
- character-specific context
- website and source-page information
- approved instructions and reference materials
- conversation history
- visitor-provided context
Each message can be processed with awareness of the workspace, character, source page, chat type, identity mode, and prior conversation.
This creates responses that are more relevant to both the user and the purpose of the experience.
- Embeddable Website Delivery-
Intari was built to operate across independent websites rather than only within a single application.A configurable WordPress plugin provides:- shortcode-based placement
- character selection
- workspace and API configuration
- visitor identity controls
- source-page capture
- multiple chat experiences across a site
- a familiar administrative interface for site owners
This separates the website presentation layer from the central conversational service.
New implementations can therefore be launched without duplicating the complete backend for every site.
- Platform Data and Future Intelligence-
The backend records structured information about each conversation, including:- workspace
- character
- source URL
- visitor identity when provided
- identity mode
- conversation type
- timestamps
- message history
This creates a foundation for future capabilities such as:
- conversation reporting
- lead scoring
- engagement analysis
- conversion tracking
- scheduling workflows
- quality evaluation
- multi-character conversations
The data model was designed not only to support current conversations, but also to make the platform more measurable and intelligent over time.
The resulting system combines product strategy, conversational design, website integration, backend services, database architecture, and configurable business behavior.
The Outcome:
Intari established a reusable foundation for deploying distinct conversational AI experiences across multiple websites and purposes.
The platform has supported implementations spanning:
- sales engagement
- coaching
- interactive demonstrations
- fictional characters
- immersive entertainment
- business-specific guidance
Each implementation can maintain its own identity and objective while using a shared technical foundation.
This reduced the need to rebuild:
- message handling
- API communication
- visitor context
- conversation storage
- website embedding
- character configuration
The platform also created a practical environment for testing how different design decisions affect conversational experiences.
These included:
- how much identity to request
- how strongly an agent should guide the conversation
- how personality influences trust and engagement
- how stage directions affect immersion
- how source-page context improves relevance
- how business outcomes can be incorporated without making the conversation feel transactional
Intari became both a working product and a framework for exploring how conversational AI can serve different human and organizational needs.
Lessons Learned:
One of the most important lessons from building Intari was that personality is not a decorative layer added after the technology is complete.
Personality affects:
- what the agent notices
- how it interprets intent
- how much it explains
- how it handles uncertainty
- how it guides the user
- whether the experience feels trustworthy, useful, or believable
A successful conversational product therefore requires both technical architecture and conversation design.
I also learned that reuse requires careful separation of concerns.
The platform should standardize infrastructure, security, data handling, and integration patterns.
It should not force every experience into the same tone, flow, or objective.
The most valuable reusable systems create consistency where consistency reduces risk while preserving flexibility where differentiation creates value.
Another lesson was that conversational AI must be designed around a purpose.
An agent that can answer almost any question may appear impressive, but it often provides less value than an agent designed to accomplish a specific task exceptionally well.
Clear purpose improves:
- prompt design
- knowledge selection
- user guidance
- measurement
- trust
The quality of the experience depends as much on what the system is intentionally not designed to do as on the capabilities it includes.
Looking Ahead:
Conversational AI is likely to evolve from isolated assistants into coordinated systems of specialized agents.
A future version of the platform could support:
- multiple characters participating in the same experience
- agents with distinct roles and areas of expertise
- handoffs between sales, support, coaching, and scheduling functions
- deeper retrieval from governed knowledge sources
- conversation quality and safety evaluation
- behavior informed by engagement history
- personalized next actions
- richer reporting for business owners
The interface may also become less visibly separate from the surrounding website.
Rather than existing as a chat window added to a page, intelligent characters could become context-aware guides embedded throughout the digital experience.
The future of conversational AI will not be defined by one assistant that tries to become everything. It will be shaped by focused intelligent experiences designed around a clear identity, trusted knowledge, a meaningful purpose, and the needs of the people they serve.
Intari was built as a foundation for that future.
Key Takeaways:
Conversational AI platforms should separate shared infrastructure from character-specific behavior and purpose.
Different use cases require distinct conversation models rather than a single generic chatbot configuration.
Personality is a functional part of product design, not simply branding.
Knowledge, visitor context, source-page information, and conversation history all contribute to relevant responses.
Configurable identity modes help balance personalization with privacy and user friction.
Centralized services and embeddable website interfaces make conversational products easier to deploy and maintain.
Structured conversation data creates the foundation for reporting, lead scoring, quality analysis, and future automation.
The strongest AI products are not those with the broadest possible capability. They are those whose architecture, behavior, and experience are aligned around a clearly defined purpose.