TalkHealth

TalkHealth

Talk instead of type. Listen instead of read. Avatars and animation to engage, educate and understand patient needs.

Tagline

Talk instead of type.

Listen instead of read.

Avatars and animation to engage, educate and understand patient needs.

Pitch

The Problem: In the UK, 1 in 3 eligible patients don't attend cervical screening when invited — roughly 1.2 million missed appointments annually. The costs cascade: each missed appointment wastes £30 in clinical time, each missed follow-up colposcopy costs £261, and a patient who never attends but develops cervical cancer faces treatment costs ranging from £1,379 (stage 1) to £19,000+ (stage 2+). Across the NHS, missed appointments consume over 1.2 million GP hours every year. The reasons are interconnected: patient anxiety (especially among women from ethnic minority backgrounds, trans patients, survivors of FGM, and deaf patients), clinical inefficiency (appointments wasted on confirming basic context), and operational waste (NHS booking slots for patients who've moved, gone private, or won't show). But beyond cost, missed screenings mean missed cancers — preventable deaths.

The Solution: TalkHealth is an end-to-end patient preparation and clinical continuity platform that addresses all three problems simultaneously. Patients log in via NHS identity and are matched to an avatar who looks like them, sounds like them, and speaks their language. The avatar explains their procedure in a calm, culturally relevant way, guides breathing exercises, and lets them submit concerns digitally — at their own pace and on their own schedule. Clinicians receive a pre-appointment brief with confirmed attendance status, patient concerns, prior clinical notes, and welfare flags — everything they need in one screen before the patient walks in. The platform integrates with appointment systems to flag unconfirmed bookings, reduce wasted slots, and manage backlogs algorithmically instead of manually.

How It Works: Patients access TalkHealth via NHS Login, select their procedure, and engage with a culturally-matched avatar who delivers clinically-adapted scripts (standard scripts for routine cases, specialized scripts for trans patients, survivors of FGM, etc.). They submit concerns and consent for the clinician to see them. Clinicians log in to a dashboard that surfaces attendance confirmation, patient concerns, prior consultation notes, and continuity links to previous visits. All interactions are stored on the patient's profile. For follow-up appointments, the next clinician picks up exactly where the last one left off — no repetition, no context loss.

What Makes It Different: Other health apps focus on patient education or appointment reminders in isolation. TalkHealth closes the loop: patient anxiety is addressed and clinical time is freed and NHS capacity is optimized in one platform. The culturally-matched avatar isn't a nice-to-have; it's the patient's entry point to structured preparation that clinicians can actually use. For procedures involving multiple consultations, continuity of care is automatic, not manual.

Impact: TalkHealth costs less than £10,000/year to run. If we recover even a fraction of the 1.2 million missed cervical screening appointments annually, the platform pays for itself many times over. A 1% improvement in attendance = 12,000 additional screenings = £360,000 in recovered wasted clinical time alone. But the real value is cancer prevention: early detection of cervical cancer at stage 1 costs the NHS £1,379 to treat; advanced stage costs £19,000+. For every patient we bring through the door, we're not just improving their care — we're preventing a disease that kills 850 women annually in the UK. For clinicians, three-minute consultations become focused on actual care instead of establishing context. For patients, especially those from underrepresented groups, they arrive seen, heard, and prepared. For the NHS, wasted capacity is reclaimed, manual data chasing is eliminated, and cancers are caught early.

Current Stage & Path Forward: TalkHealth is a functioning POC with a live demo in Indian English. We've validated the core concept with patient and clinician feedback. The next phase is clinical script approval (DSPT submission), NHS Digital integration (NHS Login, PDS API access), and expansion to 10-20 languages. The technology is proven; the real work ahead is navigating NHS governance and getting clinical sign-off so TalkHealth can scale across all healthcare procedures.

The Problem

Healthcare communication is broken at three levels, and the consequences are dire:

1. Patient Anxiety & Attendance (& Preventable Deaths) In the UK, 1 in 3 eligible patients don't attend cervical screening when invited — roughly 1.2 million missed appointments annually. The core driver is anxiety. But anxiety isn't evenly distributed. Women from ethnic minority backgrounds, trans patients, survivors of FGM, deaf patients, and those who don't speak English as a first language experience significantly worse anxiety because they feel unseen by the healthcare system. For these groups, the attendance gap is even wider.

The costs are staggering:

• Each missed cervical screening appointment wastes £30 in NHS clinical time

• Each missed follow-up colposcopy costs £261

• A patient who never attended screening but develops cervical cancer faces treatment costs from £1,379 (stage 1) to £19,000+ (stage 2+)

• Cervical cancer kills approximately 850 women annually in the UK — many of whom had never attended screening

Across the entire NHS, missed appointments waste over 1.2 million GP hours every year.

2. Clinician Inefficiency Clinicians have limited time per appointment. Much of that time is spent establishing basic context: "What's your main concern? Have you had this procedure before? Do you have any medical conditions?" Meanwhile, critical information — patient anxiety, prior consultation notes, ongoing concerns — is scattered across paper forms, separate systems, or missing entirely.

For procedures requiring multiple consultations, there is no continuity. A patient repeats their story to each clinician. Clinical notes from the previous visit aren't readily available. Context is lost.

3. NHS Capacity & Operational Waste The NHS books appointments for patients who've moved away, gone private, or simply won't show. Manual chasing by post is slow and inefficient. Appointment slots are wasted. Backlogs grow. There is no real-time visibility into which patients are confirmed attending and which are at risk of no-showing.

Data entry is manual and time-consuming. Patient concerns, submitted on paper or verbally, must be transcribed into patient records by clinical staff.

The Solution

TalkHealth solves all three problems in one integrated platform:

For Patients: Preparation & Reassurance

When a patient logs in using their NHS identity, they're matched to an avatar who looks like them, sounds like them, and speaks their language. The avatar explains exactly what's going to happen in a calm, culturally relevant way. For patients whose clinical experience is genuinely different — trans patients, those on hormone therapy, survivors of FGM — we provide clinically-adapted scripts that give them accurate information about their specific experience, not a one-size-fits-all explanation.

The patient guides themselves through breathing exercises, submits any concerns they want to share with their clinician, and consents for the clinician to see those concerns. All of this happens at their own pace and on their own schedule — whether that's the day before their appointment or the night before when anxiety peaks.

For Clinicians: Pre-Appointment Intelligence & Continuity

Five minutes before a patient walks in, the clinician opens their TalkHealth dashboard and sees:

• Is the patient confirmed attending? (Reduces no-show waste)

• What concerns did the patient submit? (Directs the conversation)

• What happened in the patient's last consultation? (Provides continuity)

• Are there welfare flags? (Alerts to patient vulnerability)

The clinician arrives prepared. The patient arrives prepared. The appointment can focus on actual care instead of establishing basic context.

For procedures with multiple consultations (e.g., colposcopy, follow-ups, treatment planning), all prior clinical notes are linked to the patient's TalkHealth profile. The next clinician picks up exactly where the last one left off. No repetition. No context loss.

For the NHS: Capacity Optimization & Waste Reduction

TalkHealth integrates with appointment systems to flag unconfirmed patients in real time. Backlogs aren't managed manually — they're managed algorithmically. Wasted slots are reclaimed. Patients at risk of no-showing get targeted support.

Patient concerns are submitted digitally and automatically linked to the patient's record. No manual data entry. No transcription errors. No paper forms losing information.

How It Works

Patient Journey:

1. Patient receives screening invitation via NHS post

2. Logs into TalkHealth via NHS Login (OIDC)

3. Selects their procedure (cervical screening, colonoscopy, blood test, etc.)

4. System retrieves patient's preferred language via NHS FHIR API (PDS - Personal Demographics Service)

5. Matched to an avatar with culturally-relevant appearance, accent, and language

6. Avatar explains the procedure step-by-step using clinically-approved scripts

7. Patient completes breathing exercises

8. Patient submits any concerns (optional) and consents for clinician to view

9. Triage responses are bundled as FHIR Observation resources and pushed to GP system

10. Patient receives appointment reminder linked to their TalkHealth profile

11. Patient attends appointment prepared and reassured

Clinician Journey:

1. Clinician logs in via NHS CIS2 (Clinical Information Services)

2. Views TalkHealth dashboard with patient list and attendance confirmation status

3. Opens individual patient record: sees submitted concerns, prior clinical notes, continuity links

4. Consultation happens with full context, focused on actual care

5. After consultation, clinician enters notes directly into the TalkHealth record, linked to the patient's profile for future reference

Real-Time Alert System (For Critical Concerns): If a patient's triage responses flag critical concerns:

1. Backend analyzes triage data and identifies critical flags

2. System bundles {patientId, reason, timestamp} into a message queue (RabbitMQ/AWS SQS)

3. Separate worker service processes the queue asynchronously

4. Worker looks up clinician contact details and sends alert (Email/Pager/EHR Alert)

5. Queue-based architecture ensures system stability even with 10,000+ simultaneous submissions

Technical Architecture

TalkHealth is built to scale with the NHS while maintaining safety, security, and clinical governance:

Frontend & Client Layer:

• React.js (TypeScript) for type safety and dynamic UI

• WebSocket (Socket.io) for low-latency, bi-directional audio streaming during avatar interactions

Backend API:

• Node.js (Express/NestJS) for high-concurrency request handling

• Handles NHS Login (OIDC) and CIS2 authentication redirects

• Orchestrates calls to NHS FHIR APIs for data retrieval and submission

AI/ML Microservice:

• Python (FastAPI) for ASR (Automatic Speech Recognition), TTS (Text-to-Speech), and RAG (Retrieval-Augmented Generation)

• ASR: IndicASR and HuggingFace models for multi-language speech recognition

• TTS: IndicTTS and HuggingFace models for natural, accent-matched speech synthesis

• RAG Pipeline:

  • Ingestion: Medical documents chunked and embedded into vector DB

  • Retrieval: User questions retrieve top-3 relevant chunks

  • Guardrail: Similarity score threshold (≥0.75) ensures AI only answers medical questions with high confidence. Below threshold → "Please contact your GP"

  • System Prompt: "You are a helpful assistant. You MUST ONLY answer based on the provided context. If the answer is not in the context, say you do not know."

  • This architecture mathematically prevents the AI from answering off-topic questions (e.g., car repairs, unrelated diseases)

Databases:

• PostgreSQL (Primary Operational DB): Stores triage results, interaction logs, CMS data for testimonials, and temporary state

• Pinecone or pgvector (Vector DB): Powers RAG retrieval for medical knowledge base

• NHS Systems (via FHIR Interface): Read/write to Electronic Health Records (EHR) using HL7 standards

Message Queue:

• RabbitMQ or AWS SQS for asynchronous processing of critical alerts

• Decouples alert generation from delivery, ensuring the main API never times out or crashes under high load

• Messages persist until processed, guaranteeing no alert is lost

Authentication & Identity:

• Patients: AWS Cognito or Keycloak redirects to NHS Login (OIDC). Returns NHS Number in token.

• Clinicians: Redirects to NHS CIS2 for staff authentication

• Both flows use industry-standard OIDC protocols for NHS integration

NHS Integration (FHIR/HL7):

• Data Retrieval: Using NHS Number, backend calls NHS FHIR API (PDS) to fetch preferred_language, demographics, and existing medical history

• Data Submission: Triage responses bundled as FHIR Observation or QuestionnaireResponse resources and pushed back to GP system

• Interoperability: Full HL7 compliance ensures TalkHealth can integrate with any NHS trust's EHR

Design Principles:

• Local-First Operations: Temporary triage state and testimonials stored locally in PostgreSQL; only synced to NHS systems when necessary. This reduces latency and load on national infrastructure.

• Asynchronous Processing: Critical alerts queued and processed by dedicated workers, preventing API bottlenecks

• Language Decoupling: Question IDs stored once; content returned dynamically in requested language (Hindi, Urdu, Polish, etc.)

• Safety by Design: RAG guardrails, similarity thresholds, and hardcoded fallbacks ensure AI never provides unsafe medical advice

What Makes It Different

Other health apps solve one problem:

• Appointment reminder apps improve attendance slightly but don't address anxiety or clinical continuity

• Patient education portals provide information but don't integrate with clinical workflows

• Clinical dashboards show patient data but don't include pre-appointment patient preparation

TalkHealth closes the loop. The culturally-matched avatar isn't a comfort feature — it's a clinical preparation tool that generates structured data (patient concerns, engagement level, anxiety triggers) that clinicians can actually use. Patient anxiety is addressed and clinical time is freed and NHS capacity is optimized in one platform.

For procedures with multiple consultations, continuity is automatic, not manual. Patients don't repeat themselves. Clinicians don't lose context.

On the technical side: Unlike generic chatbots, TalkHealth is built with NHS standards from day one. FHIR integration, CIS2 authentication, PDS data retrieval — these aren't afterthoughts, they're the foundation. The RAG guardrails ensure medical safety. The queue-based architecture ensures operational reliability at NHS scale. We're not building a health app and then trying to shoehorn it into the NHS. We're building for the NHS from the ground up.

What We've Built & What We've Learned

What We've Built (POC):

• A fully functional, interactive avatar demo in Indian English that patients can engage with

• Demonstrated culturally-matched avatar generation using LatentSync on Vertex AI

• Designed the complete architecture for NHS integration (FHIR APIs, PDS, CIS2)

• Conceptualized RAG guardrails, queue-based alert systems, and clinical continuity workflows

What We've Learned (Building the Avatar):

Technical Challenge: Getting avatar generation to work on current infrastructure was harder than expected. SadTalker (the initial avatar library) was built for older versions of Python, NumPy, and torchvision. Running it on current Colab infrastructure required patching four separate source files just to generate a single video. This taught us something crucial: open-source models built for legacy environments don't scale in production.

Our Solution: We replaced SadTalker with LatentSync on Vertex AI. LatentSync is actively maintained, purpose-built for avatar generation at scale, and runs natively on modern infrastructure. This decision informed our broader architectural philosophy — we prioritize production-grade, actively maintained dependencies over cutting-edge-but-fragile libraries. For TalkHealth at NHS scale, reliability and clinical safety trump experimental features.

Clinical & Governance Insights: Through early conversations with clinicians and patient advocacy groups, we learned that the real bottleneck to scaling healthcare tech isn't the technology — it's NHS approval. Clinical script approval, DSPT compliance, NHS Digital integration approvals, and continuity governance processes take 6-12 months. This isn't a problem to work around; it's a feature. These processes exist to protect patients. Our commitment is to build with NHS processes from day one, not against them.

Next Steps & Implementation Roadmap

Immediate (Next 3 Months):

• Build the patient-facing frontend (React.js, TypeScript) with NHS Login integration

• Develop the Node.js backend API with FHIR/HL7 interfaces

• Expand avatar library from 1 to 10-20 languages (Hindi, Simplified English, Polish, Urdu, Mandarin, etc.)

• Design clinically-adapted scripts for specialized patient groups (trans patients, FGM survivors, etc.)

• Conduct clinical script review workshops with gynecologists and patient safety teams

How We'll Improve & Scale It:

• Multi-language expansion: Test avatar effectiveness across different languages and accents; gather feedback on cultural relevance

• Clinical integration: Build FHIR APIs for EHR integration; work with NHS trusts to test with real appointment systems

• AI safety: Implement RAG guardrails with similarity thresholds; establish hardcoded fallbacks for unsafe medical questions

• Dashboard development: Build the clinician dashboard with attendance confirmation, patient concerns, and continuity links

• Message queue system: Implement RabbitMQ/SQS for asynchronous alert processing at scale

• User experience testing: A/B test breathing exercises, script delivery methods, and dashboard layouts with real patients and clinicians

Medium Term (6-12 Months):

• Complete backend API and database architecture (PostgreSQL, vector DB)

• Achieve NHS Digital approval for NHS Login and PDS integration

• Complete DSPT certification

• Launch cervical screening pilot with 3-5 NHS trusts

• Expand to colonoscopy and other high-anxiety procedures

• Establish baseline metrics: attendance rates, clinician time savings, patient satisfaction

Long Term (12+ Months):

• Scale to all NHS healthcare procedures (blood tests, imaging, surgery, mental health consultations, etc.)

• Prove the impact: quantified attendance improvements, clinician time reclaimed, NHS capacity optimization

• Expand beyond NHS to private healthcare, international health systems

• Continuously update clinical scripts and avatar training based on new medical evidence

Impact

Financial Impact: TalkHealth costs less than £10,000/year to run. If we recover even a fraction of the 1.2 million missed cervical screening appointments annually, the platform pays for itself many times over.

A 1% improvement in attendance = 12,000 additional screenings recovered:

• £360,000 in recovered wasted clinical time (£30 per appointment × 12,000)

• Early detection of stage 1 cervical cancer (£1,379 per case) instead of advanced stage (£19,000+ per case) = potential savings of £200M+ annually across the NHS if scaled nationally

Clinical Impact:

• Patients arrive prepared and reassured, especially those from underrepresented groups

• Clinicians save several minutes per appointment by having context upfront (across 1000s of appointments, this is significant capacity reclaimed)

• No more lost context in multi-consultation procedures

• Better clinical outcomes because consultations are focused on actual care, not information gathering

Equity Impact: Women from ethnic minority backgrounds, trans patients, survivors of FGM, deaf patients, and those who don't speak English as a first language feel seen and understood by the healthcare system for the first time. Attendance gaps narrow. Early detection improves. Health inequities shrink.

Life Impact: Cervical cancer kills approximately 850 women annually in the UK. Many of these deaths are preventable with early detection. For every patient TalkHealth brings through the door who would otherwise have missed screening, we're potentially preventing a life-threatening diagnosis. That's not a metric. That's why we're building this.

Operational Impact: NHS capacity is no longer wasted on no-shows, manual chasing, and data entry. Backlogs are managed algorithmically. Clinicians are freed to focus on care.

TalkHealth isn't another generic health app. It's a platform that sees every patient as an individual and makes that visibility useful to clinicians and operational teams. That's what makes people show up — and what saves lives.