The work behind the system.
A website assistant needs to handle an ongoing conversation, staff takeover and unavailable replies as well as the answer itself. This internal implementation documents the integration present in the Keen Systems website. It does not claim a measured business outcome or a deployment for an external customer.
Explore Keen AI and its supported website experienceWhat the project brings together.
Context-aware reply requests
The website sends the latest enquiry and up to 30 recent conversation messages to the configured Keen AI reply service. A reply is shown only when the upstream service returns the expected reply decision.
Human handover
A team-member request pauses AI replies, marks the chat for attention and supersedes pending reply work. The visitor can also continue with the team on WhatsApp; staff response depends on availability.
Conversation continuity
Before posting an answer, the integration checks for newer messages, closed chats and staff assignment. Selected temporary failures have bounded retries and an understandable fallback message.
The workflow, simply.
- Visitor enquiry
- Recent context & reply request
- Current-chat checks
- AI reply or human continuation
Architecture overview.
The Next.js website stores chat sessions, messages and queued AI reply work in MySQL.
A server-side authenticated request sends the enquiry and recent context to a configured Keen AI service; credentials are not exposed to the browser.
Queued work uses leases and final transaction checks to avoid inserting a stale reply after the conversation has changed.
Human handover pauses automation and exposes WhatsApp continuation. The website integration alone does not establish the upstream model's training or knowledge sources.
Checks for a comparable deployment.
- Send a follow-up while an AI reply is in progress and verify that the older reply is superseded.
- Request a team member and confirm pending AI work does not continue to post answers.
- Test staff takeover, closed chats and an unavailable upstream service.
- Review answers against approved information and verify the visitor can reach a person when required.
Useful things to know.
Is this an external customer project?
No. This is the internal Keen Systems website assistant integration. Its implementation is described without invented customer names, conversion figures or response-time claims.
Does the implementation train a custom model?
The website calls a configured reply service with recent conversation context. That is an integration, not evidence of custom model training. Model training and private knowledge integrations would require a separately defined scope.
Can a similar assistant connect to business software?
A comparable project can be scoped for approved CRM, ERP or booking integrations. Each connection needs defined permissions, record ownership, error handling and review before it can make business changes.
