NexavateAI Trust Center
AI Transparency
We believe healthcare enterprises deserve to know exactly which AI models power NexavateAI, what data each model sees, how human oversight operates, and the clear boundaries of what our AI will and will not do.
Our AI Commitment
Core principles governing AI use in NexavateAI
AI assists, humans decide
AI in NexavateAI helps with scheduling, information retrieval, and workflow support. All clinical decisions remain with qualified healthcare professionals.
Transparent by design
We name the models we use, describe what data each model processes, and are clear about what AI cannot do. No hidden AI features.
AI Products
Two AI-powered products with distinct user audiences
SYRA — Patient Assistant
A conversational AI assistant on the clinic's public website. Helps patients find clinic info, understand services, book appointments (with OTP), and get answers to common questions.
- Answers patient queries using clinic knowledge base
- Guides appointment booking via structured conversation
- Escalates complex or medical queries to human staff
- Does not provide medical advice or diagnoses
NEXA — Clinic Copilot
An internal AI copilot for clinic staff. Assists with appointment management, patient record retrieval, document summarisation, and prescription drafting.
- Retrieves and summarises patient records
- Assists prescription drafting (reviewed by doctor)
- Manages slot availability & scheduling
- Does not make final clinical decisions
Models Used
Named AI models powering each feature
| Model / Service | Provider | Used by | Purpose |
|---|---|---|---|
| GPT-4o / GPT-4o-mini | OpenAI | SYRA, NEXA | Primary conversational AI — answering queries, intent classification, booking guidance, document summarisation |
| text-embedding-3-small | OpenAI | SYRA, NEXA | Creating vector embeddings of clinic knowledge base for semantic retrieval (RAG) |
| whisper-large-v3 | Groq | SYRA (optional) | Speech-to-text transcription for voice input (when voice feature is enabled) |
Model versions may be updated to newer stable releases as providers improve accuracy and safety. This page will reflect current versions.
What Data AI Sees
Precise description of inputs passed to each model
| AI feature | Data the model receives | Data NOT sent to AI |
|---|---|---|
| SYRA patient chat | Patient's chat messages (current session), clinic knowledge base (services, timings, FAQs), current booking state | Other patients' data, previous session history, authentication credentials, raw OTP codes |
| NEXA staff copilot | Staff's natural language query, retrieved appointment/patient records, prescription templates | Other clinic's data, raw database credentials, system secrets |
| Embeddings (RAG) | Clinic knowledge base text segments (clinic docs, FAQs) | Personal patient data, appointment records, financial data |
| Speech transcription | Audio segment recorded during voice session | Other sessions' audio, authentication data |
Human Oversight
Control mechanisms ensuring humans stay in the loop
Appointment confirmation
All appointments booked via SYRA are visible to clinic staff in the admin dashboard for review and confirmation.
Prescription review
NEXA-assisted prescriptions are drafts only. A doctor must review, edit if needed, and explicitly confirm before any prescription is finalised.
Slot control
Clinic staff manually control which time slots are available. AI cannot block or open slots without explicit staff action.
Intent boundaries
SYRA is configured with system-level instructions that restrict it from providing medical diagnoses or treatment recommendations.
Escalation path
When SYRA cannot answer a query (medical, emergency, complex), it instructs the patient to contact the clinic directly or call emergency services.
No autonomous actions
AI does not initiate outbound communications, modify records, or perform actions without a human-triggered event. All AI responses are reactive, not proactive.
Human Review Required
Mandatory human authority for clinical and operational decisions
NexavateAI is designed so that AI assists workflow — it does not replace qualified human judgment. The following actions require explicit human review or authorization before they take effect:
| Action | AI role | Human requirement |
|---|---|---|
| Prescription generation | NEXA may draft prescription content based on staff input | A licensed clinician must review, edit if needed, and explicitly confirm before any prescription is finalised |
| Clinical decisions | AI does not diagnose, treat, or recommend specific clinical interventions | All clinical decisions remain with qualified healthcare professionals |
| Appointment modifications | SYRA may assist patients with booking requests | Cancellation, rescheduling, and slot blocking require authorised clinic staff or verified patient action through defined workflows |
| Slot availability | AI reads current slot state | Only authorised users may block, unblock, or modify clinic schedule availability |
| Patient record changes | AI may retrieve and summarise records for staff | Record modifications require authenticated staff action; AI does not autonomously alter records |
Principle
AI assists. Humans decide. No AI output in NexavateAI constitutes a final clinical, legal, or operational decision without human review where applicable.
Known Limitations
What AI in NexavateAI cannot do reliably
| Limitation | Implication | Mitigation |
|---|---|---|
| Hallucination risk | AI may produce plausible but inaccurate responses | Knowledge base grounding (RAG), human review of clinical outputs |
| No real-time data | AI knowledge is based on the clinic knowledge base, not live external sources | System prompt scoped to clinic context; AI refers patients to staff for real-time info |
| Language variability | AI performance may vary with non-standard spellings, regional dialects | Input normalisation; primary testing in English; multilingual support in roadmap |
| Context window limits | Very long conversations may lose early context | Structured booking state machine reduces dependence on long context |
| External model dependency | Availability dependent on OpenAI / Groq service uptime | Graceful degradation messaging; fallback prompts on API errors |
What AI Does Not Do
Hard boundaries — by design and by policy
Never does
- Provide medical diagnoses or suggest specific treatments
- Replace a doctor's clinical judgment
- Access patient data across different clinics
- Initiate communications without human instruction
- Make final decisions on prescriptions or care plans
- Store raw OTP codes or authentication credentials
- Use patient data to train foundation models
Always does
- Refer emergency situations to emergency services
- Stay within the clinic's configured knowledge boundary
- Disclose that it is an AI assistant when asked directly
- Restrict scope to scheduling & operational queries
- Escalate complex medical questions to human staff
- Respect session isolation — no cross-session data leakage
Training & Data Retention
How our AI use relates to model training
Our policy
NexavateAI does not use clinic or patient data to train AI models. Data sent to OpenAI or Groq APIs is governed by their API data usage policies. NexavateAI operates under OpenAI's API terms, under which OpenAI does not use API inputs to train models by default.
Clinic knowledge base embeddings (clinic docs, FAQs) are stored in Qdrant Cloud as vector representations only — not as raw text. This data is used exclusively for retrieval to improve AI response accuracy for that specific clinic.
Patient chat messages, booking data, and personal information are not vectorised, embedded, or used in model training pipelines.
AI Questions & Concerns
If you have questions about how AI works in NexavateAI, want to understand a specific interaction, or have concerns about AI safety in a healthcare context, please reach out.
NexavateAI — AI & Trust
Email: info@nexavateai.com