Changelog

Updated the short preview clips for eight AI patient cases with more natural, consistent movement and improved visual quality.

Hover over Susan’s portrait in the homepage patient panel to see her AI Actor preview before starting a consultation.

Added a prominent “Meet all patients” link from the homepage so you can jump straight to the AI Actors module.

Added a paid mock exam add-on with unseen stations, timed reading and consultation phases, resumable attempts, per-station marking, and a final examiner-style report.

Cases 51-55 are now live, bringing the AI Actor library to 55 cases with new direct-voice roleplays and updated case metadata across the app.

The AI Actors hub, dashboard, homepage, and My Progress now surface the mock exam and progress next steps more clearly.

Realtime voice access checks and session startup handling were tightened so eligible consultations start more consistently.

Added Case 47 to Case 50 to bring the AI Actor library up to 50 cases, covering PCOS, sciatica and red-flag screening, recurrent UTI prevention, and erectile dysfunction with cardiovascular risk.

Cases 47-50 use a dedicated browser-to-Deepgram speech-to-text stream, GPT-5.5 patient simulation, and ElevenLabs speech playback, with short-lived Deepgram tokens minted server-side after AI Actor access checks.

Expired credit grants are now marked once they have been processed, so future balance checks do not repeatedly deduct the same expired credits.

The site Content Security Policy now allows Deepgram HTTPS and WebSocket connections, and temporary Deepgram tokens are passed with the correct bearer WebSocket auth scheme.

Classic, Quick, SCA, Quiz, and case-review tutor responses now point at GPT-5.5 by default. WhatsApp tutor responses remain on GPT-5.4.

Both web and mobile AI marking now use the shared GPT-5.5 marking model configuration for domain scores, feedback, and improvement priorities.

The homepage, AI Tutor page, pricing copy, AI Actors surfaces, FAQ, help guide, account WhatsApp section, and public AI metadata now call out the correct model where relevant.

A compact mode pill in the chat input lets you switch between Quick, SCA, and Quiz modes. Each mode adjusts response length, structure, and conversation style to match your revision goal.

Type / in the chat input to access six exam-focused templates: /explain, /differentials, /management, /red-flags, /safety-net, and /patient-friendly. Each shapes the AI response around a specific SCA skill.

Save any AI Tutor message as a flashcard with one tap. A dedicated Flashcards page lets you flip through saved cards, edit them, and delete ones you no longer need — ideal for spaced repetition before the exam.

After marking a roleplay, "Review this attempt with AI Tutor" opens a new chat with server-loaded case, mark scheme, score, weakest domain, and saved feedback context. Saved attempts in Past Attempts can be reopened later too.

AI Tutor now uses your recent roleplay performance — including weaker domains and cases to revisit — to tailor answers when it is relevant.

The suggested starter prompts now change to match the selected tutor mode, so you always see relevant entry points whether you are in Quick, SCA, or Quiz mode.

The chat shows when it is reviewing a specific case attempt, the tutor input shows a Personalised badge once progress exists, and My Progress includes a direct action to practise the weakest domain with AI Tutor.

Added Case 42 to Case 45 to broaden the roleplay library with older adult, urgent care, and undifferentiated presentations. The new cases cover GCA red flags in suspected PMR, exertional chest pain and stable angina, painless visible haematuria, and a proxy-call TIA scenario.

Cases 42-45 now include structured Learn content, key clinical topics, consultation approaches, common pitfalls, and case-specific Consultation Coach mark schemes with critical actions.

Cases 21-41 now use the newer consultation structure with Consultation Coach support, clearer inspector tabs, and more consistent Doctor Notes and My Notes labelling.

The new case pages now use a clearer Reason for Appointment label and include Recent Results sections, so the opening clinical context and investigation data are easier to scan before or during practice.

The AI Actors index, category filters, progress tracking, agent mappings, and case metadata have been updated so the full 45-case library appears consistently across the app.

The AI Actors header now suggests a focused next practice step based on your progress, streak, and recommended cases, with a direct link to pick today's case.

The chat header navigation now points users back to the main chat page, keeping the route behaviour clearer and more predictable.

AI model names and supporting colour treatments were tidied across product surfaces, FAQs, marking routes, and provider configuration.

The homepage, welcome page, how-it-works section, premium CTA, and Learn view copy received layout, colour, and clarity improvements.

A new hero-slider panel and updated How It Works image were added to give the public site more variety and clearer product context.

The footer now includes sister-platform links, and trainee messaging across the site was updated from 1,000+ to 1,200+ trainees.

Credit package information was standardised across the homepage, purchase flow, Stripe webhook handling, and upgrade modal.

Case controls and the first twenty AI Actor case pages received small layout and accessibility refinements for a more consistent practice experience.

The AI Actors page gained quick-buy credit packages, giving users a faster path to add credits from the practice flow.

Upgrade prompts, credit usage text, package descriptions, and the upgrade modal feature list were rewritten for clarity and accessibility.

Cases 8-20 were refactored into the newer case layout, with improved structure, cleaner interaction patterns, and expanded Consultation Coach registry coverage.

Header and hero slider images were tuned for better loading behaviour, helping key visual elements appear sooner.

Header button text and styling were refactored into a shared pattern so the logged-in and logged-out headers behave more consistently.

Coach now reads the live transcript and surfaces three kinds of cue inline as the patient speaks:

Each cue shows a short clinician's mental note up top, and expands to show the verbatim patient quote it's grounded in. Cues are deliberately rare — Coach prefers silence to noise.

Coach is opt-in, off by default, and remembers your preference per browser. Toggle it on once and it sticks — toggle it off and Coach goes silent and stops making any background requests.

All seven AI Actor case pages have been restructured around a new three-column layout: the patient on the left, the live transcript in the centre once the consultation starts, and your Doctor Notes and Coach panel on the right. The consultation controls (start/end/mute) are now lifted into the page chrome so they stay visible no matter which inspector tab you're on.

The Practice ↔ Learn switch on each case has been redesigned as a compact pill toggle that takes up less vertical space and adapts better to small screens.

When you run out of AI Actor credits, the upgrade dialog now hides the Revision Essential plan (which doesn't include credits) and surfaces inline credit top-up packs (3, 10 or 25 credits) so you can keep practising without leaving the page.

Coach runs on a separate, fast model (gpt-5.4-mini) so cues appear within a second or two of the patient finishing a sentence — without slowing your main AI Tutor or marking model.

Every cue Coach surfaces is validated against the live transcript on the server — if the quote isn't a verbatim substring of what the patient actually said, it's rejected. You'll only ever see cues grounded in real evidence.

The upgrade dialog now uses dynamic viewport units so the full plan grid is reachable on mobile browsers (no more cut-off CTA at the bottom). The duplicate close button has also been removed.