KEEMOV Intake AI

Voice-first AI intake that helps fitness and rehabilitation professionals collect structured, reviewable client information before the first assessment.

Comment

Inspiration

Small fitness, rehabilitation and physiotherapy practices often collect initial information through static forms, messaging apps or an unstructured conversation at the start of the first appointment. That creates two costs: valuable professional time is spent transcribing and reorganizing information, and relevant details about goals, previous treatment, symptom behaviour, limitations or warning signs may be missed. KEEMOV Intake AI was created to make this first step more consistent and accessible without replacing professional judgement.

What it does

A professional creates an intake session and sends a secure web link. The client completes a voice-first interview with an AI assistant. The interview adapts to prior answers and follows a constrained professional workflow rather than behaving like an open medical chatbot. When the interview is complete, KEEMOV produces a transcript, structured anamnesis data, a professional summary, non-diagnostic risk signals, suggested follow-up questions, recommended functional tests, dashboard insights and a PDF report. The professional reviews, validates and decides how to use every output.

How AI operates the product

Gemini is live in the core workflow. Gemini Live handles the real-time voice conversation and produces conversational turns. A deterministic clinical planner and completeness rules constrain which information blocks are still missing and keep the interview focused. Gemini 2.5 Flash then transforms the transcript into structured data and professional-facing report content. AI therefore creates the central service - an adaptive interview and a structured preparation document - while the human professional remains responsible for interpretation, assessment and decisions. The system does not diagnose or prescribe treatment.

How we built it

KEEMOV Intake AI is a Next.js and TypeScript web application deployed on Vercel. Firebase provides authentication and application data services. Gemini API and Gemini Live API power the AI workflow. Stripe provides subscriptions, tax calculation, customer self-service and webhook-driven entitlements.

The first repository commit was created on May 29, 2026, after the hackathon start. Development proceeded through small branches, pull requests, automated tests and a release gate. By August 14 the repository included thousands of automated tests and a production-validated billing workflow.

Challenges we ran into

The hardest problem was not simply connecting a language model. A useful voice intake must know what is missing, avoid endless conversation, recover safely from interruptions, produce consistent structured output and remain clearly outside diagnosis. We therefore combined Gemini with deterministic planning, lifecycle controls, retry and stale-analysis recovery, quota enforcement and professional review. A second challenge was responsible handling of sensitive information. The current external pilot is restricted to synthetic data while consent, retention and deletion requirements are completed.

Accomplishments that we're proud of

We are proud that KEEMOV Intake AI progressed beyond a chatbot prototype into a working production workflow. A professional can create a secure session, a synthetic client can complete an adaptive Gemini Live interview, and the system produces structured anamnesis data, follow-up questions, functional-test suggestions and a reviewable PDF report.

The product also includes deterministic planning, completeness controls, interruption and stale-analysis recovery, quota enforcement, production billing and thousands of automated tests. Most importantly, it is designed to amplify professional judgement rather than replace it: AI structures and proposes, while the professional validates and decides.

What we learned

We learned that usefulness comes from the boundary between AI flexibility and deterministic control. Voice makes the experience natural, while completeness criteria, explicit lifecycle states and professional validation make the output dependable.

The pilot also identified a necessary privacy improvement: before a voice interview begins, the client should see a clear data-processing notice and provide the required acknowledgement or consent. This will be implemented together with retention and deletion controls before KEEMOV Intake AI is authorized for real-patient workflows.

What's next for KEEMOV Intake AI

The next product phase will introduce a clear data-processing notice and the required acknowledgement or consent before every interview, together with retention and deletion controls. We will also make specific functional-test recommendations more prominent in the professional report.

After these privacy and usability improvements are validated, we will expand supervised professional pilots and evaluate repeated use, retention and real unit economics before authorizing real-patient workflows. Future integration with KEEMOV Assess will connect the intake report with guided functional assessment.

Built With

Share this project:

Updates