Information changes.
Work continues.
Prices, programmes and schedules need review. Customer messages arrive while replies are being prepared. Some requests need staff confirmation.
IMFIT / Owner-operated implementation
How IMFIT connects
AI, knowledge and staff work.
Connecting customer enquiries, approved business knowledge, staff follow-up and mobile software inside a fitness business.
Try the interactive demoThe answer is only
the beginning.
A customer asks a question. The business has to use current information, recognize what is still unknown, and make sure the next step reaches a responsible person.
IMFIT brings these concerns into one software system. AI supports interpretation and communication. Business rules handle state, approvals, reporting and the controls around the response.
Information changes.
Connect the answer
to its context.
Version approved knowledge, preserve conversation state, check before applying a result, and make handoff part of the application.
IMFIT / Implementation walkthrough
A reply is useful only
if it is still the right reply.
IMFIT is our own operating business. Its software connects enquiries, business knowledge and staff controls. Here are the mechanisms you can inspect—not a promise that every business needs the same system.
Illustrative enquiry
“What are your opening hours?”The question is simple. The information and conversation can change while the answer is being prepared.
- 01 / Establish context
Find approved information.
Knowledge is filtered by approval status and effective dates. Source identifiers and versions remain available in the response context.
Inspected source + selected local tests - 02 / Check freshness
Has another message arrived?
A transaction compares the expected inbound message with the latest one. An older result is rejected when they no longer match.
Inspected source; demo reconstructs the check - 03 / Respect staff control
Should automation still send?
The outbound gate checks human takeover and persistent opt-out. Assigning a staff member alone does not mean takeover.
Inspected source + selected local tests
Try the mechanisms yourself.
Change a knowledge version, prepare a reply and test an interruption. The public demo uses fictional records and deterministic responses; it does not connect to IMFIT production or a live model.
This sequence explains inspected controls. It is not an end-to-end production certification. The demo’s unknown-policy review task is illustrative; no identical production review queue is claimed.
How the pieces connect
More than a chat interface.
An implementation map based on the inspected IMFIT code and workflow documentation. The demo above isolates three of these mechanisms.
Customer input
Enquiry & conversation
A question, a preference, a request for staff.
Business context
Approved knowledge
Current facts, source references and versions.
Application layer
Interpret. Check. Coordinate.
AI-assisted replies work alongside explicit state, permissions and business rules.
Customer response
Use relevant context; handle unknowns.
Staff follow-up
Keep ownership and the next step visible.
Operational software
Mobile access, shared records and reporting.
Engineering evidence
Specific mechanisms.
Inspectable behavior.
The strongest evidence is in how the application handles changes, missing information and interruptions.
Only usable knowledge reaches the reply context.
The inspected availability check excludes drafts, rejected and superseded facts, and facts outside their effective dates. Structured knowledge carries an ID and version into reply context.
Verified locally: approval/date availability and structured source-reference tests passed.
A newer message invalidates an older result.
Before updating a lead, a Firestore transaction compares the expected inbound message ID with the latest stored message ID. A mismatch returns a superseded-reply error.
Verified by source inspection. The demo exercises the comparison; it does not reproduce a live Firestore transaction.
Staff takeover pauses automated replies.
The inspected outbound gate recognizes human takeover and persistent opt-out. An ordinary staff assignment alone is not equivalent to takeover.
Verified locally: takeover/opt-out gate and opt-out-priority tests passed.
5 selected backend tests passed
Executed against the local IMFIT checkout on 30 October 2026. This verifies those specific assertions, not model quality, production delivery or commercial impact.
The application
around the intelligence.
The repository spans more than message generation: a Flutter mobile application, Go backend services and Firebase-based data and authentication.
Mobile & backend
Application components support members, leads and daily operations through shared records and services.
Operational priorities
Staff call priorities use explicit scoring rules. They are not calibrated probabilities that a lead will convert.
Management reporting
Attendance and revenue reporting use deterministic data processing. The arithmetic does not depend on a model’s generated text.
Evidence and next questions
Show the capability.
Measure the outcome.
IMFIT is an owner-operated implementation, not an independent customer endorsement. Source and selected tests establish specific mechanisms; they do not establish conversion, retention or revenue uplift.
Visit IMFITWhat to measure next
- Answer quality: corrections and source support on representative enquiries.
- Handoff: whether staff receive enough context and complete the next step.
- Operations: handling time and staff usage compared with a baseline.
- Economics: model and support cost per useful completed task.
The case does not establish an MCP deployment or an autonomous agent. Those remain separate capabilities to implement and evaluate.
Where this experience transfers
IMFIT is an owner-operated engineering reference. Its knowledge, workflow and application patterns inform our work on new client projects. It is not a ready-made fitness product; every new application and integration needs its own scope and evaluation.
Discuss an application featureScoped AI engineering
Have a task in mind?
Let’s assess the fit.
Start with the work, the information it needs and the question you want to answer.