AI-native product · HealthTech · iOS, Android and web
Cognic AI. Less work after a session, full therapist control.
We designed and built an AI assistant that turns a session recording into a transcript, summary, and working note. The practitioner reviews the output, edits it, and decides what becomes part of the final documentation.
Post-session workflow
- 01Recording
- 02Speaker-labelled transcript
- 03Draft for review
A product built around the practitioner’s real work
After a meeting, a therapist needs to reconstruct the conversation, organise the important information, and prepare documentation. With several sessions a day, that work accumulates, consumes time, and increases the risk of missing important context.
Cognic was created as a specialised post-session assistant—not as a chatbot that conducts therapy and not as a system that makes clinical decisions.
The challenge: automate the work without handing responsibility to AI
In this product, speed could not outrank trust. Recordings and notes contain sensitive context, models can miss nuance, and generated text may sound more certain than the source material allows.
We had to combine a simple mobile flow, reliable transcription, flexible note templates, and manual correction. At the same time, the project required legal and data-protection input and a product foundation ready for multiple markets.
The solution: working material that stays under human control
Cognic records a session in the mobile app, separates speakers, and then prepares a transcript, summary, and note draft in the chosen format. The user reviews the material, corrects it, and approves the final content.
Patient history supports continuity, while the web panel makes longer review and editing easier on a larger screen. For clinics, the product supports organisations, invitations, and a shared usage pool.
04 / Workflow
From a completed session to a structured note
The interface guides the user through one clear process. AI does the preparatory work, but the final step always belongs to the practitioner.
- 01
The therapist records a session
A single action starts recording in the mobile app. The product communicates its scope and supports an informed-consent process.
- 02
The system structures the conversation
The recording becomes a speaker-labelled transcript, making it easier to return to the right part of the conversation.
- 03
AI prepares working material
Cognic creates a summary and note draft in the selected format, including a structured SOAP note.
- 04
The practitioner verifies the result
The therapist compares the draft with the transcript, edits the content, and decides what belongs in the final documentation.
05 / Scope
The product goes beyond a single summary
What began as a tool for one practitioner evolved into a connected mobile and web ecosystem for therapists and clinics.
Speaker-labelled transcription
A readable conversation record makes context easy to find and the AI draft easier to verify.
Summaries and note templates
Users choose how to structure the material, including the SOAP format.
Patient history
Previous materials are available in one place to support continuity of work.
Mobile app and web panel
Recording happens on mobile, with longer review and editing also available on a larger screen.
Clinics and teams
Organisations can invite users and work with a shared usage pool.
A multilingual product
The interface and summaries support Cognic’s growth beyond Poland.
A product built around the real session workflow
Official Cognic screens show the flow from recording through transcription and summary to an organised work history.
Responsible AI
AI supports documentation. It does not replace clinical judgement.
The key product decision was to treat model output as a draft, not as unquestionable documentation. This deliberately keeps a human in the loop and limits automation risk.
We developed the solution alongside legal and data-protection consultations. Product communication does not promise diagnosis, therapy, or error-free AI output.
- AI output always requires practitioner review
- the user decides the final content of the note
- the product clearly separates documentation from diagnosis and therapy
- consent, retention, and recording scope are part of the workflow—not an afterthought
06 / Delivery
How we developed Cognic
We started with technology and legal risks, then expanded the feature set and product distribution.
01. Model and workflow validation
We compared online models with on-device approaches, evaluating transcription, summary quality, and data-protection options. In parallel, we prototyped a simple flow: start a session, process it, and review the result.
02. Beta with therapists
We began with internal testing and then developed the product using practitioner feedback. One thousand therapists expressed interest during the first month of beta testing, helping us prioritise features and simplify the daily workflow.
03. Mobile, web, and teams
Later stages introduced the web panel, patient history, note templates, and organisation features. The architecture and commercial model now support both individual practices and clinics.
04. Global readiness
We aligned the interface, pricing, and communication across five languages and expanded store distribution. The same product can now be localised without building separate applications for every market.
07 / Stack
Product architecture
The stack supports fast development across two mobile platforms, AI processing, and continued product scaling.
- React Native + Expo
- one mobile product codebase for iOS and Android
- Node.js
- server-side application logic and process orchestration
- PostgreSQL
- structured product and user data
- Redis
- fast task processing and temporary data
- Speech models and LLMs
- transcription, speaker separation, summaries, and note drafts
- RevenueCat
- consistent subscriptions and packages across the App Store and Google Play
08 / Outcome
The outcome: from an idea to a multi-market product
Cognic progressed from AI validation to a working ecosystem for individual therapists and teams.
- 011,000 interested therapists during the first month of beta testing
- 02an average of 30 minutes less post-session work, based on beta-stage data
- 03an iOS and Android mobile app supported by a web panel
- 04five language versions: PL, EN, DE, ES, and IT
- 05global distribution across 175 App Store regions and 173 Google Play regions
- 06a product model for both individual practices and therapy clinics
“The team was a great partner in the project—punctual, engaged, and flexible. We felt they genuinely cared about the final result. The app met our expectations, and communication was smooth at every stage.”
A global market without losing local context
Cognic is ready for worldwide distribution, but each market still requires local communication, legal review, and workflow fit. The product therefore combines a shared technology core with market-specific languages, pricing, and content.
That approach scales the solution without copying the application, while recognising that international expansion requires more than translating an interface.
What this project demonstrates
Cognic demonstrates how to build AI for a high-trust environment: start with a specific user task, design human control into the workflow, and only then scale models, features, and markets.
For GMI, this was end-to-end product engineering—from AI experiments and the mobile app through backend and web to subscriptions, localisation, and global distribution.
Have a workflow that AI could genuinely lighten?
Start with one workflow, not a sweeping promise.
In 10 days, we will select a use case, examine the data and risks, and prepare a practical AI pilot plan.