Built on Proven Frameworks
Every Howie-AI product is grounded in clinical evidence, regulatory standards, and production-grade technical architecture. This page is for councils, commissioners, procurement teams, and technical evaluators who need the detail.
NHS & GOV Framework Alignment
Howie-AI products are designed with direct reference to the following frameworks and guidelines.
| Framework | Full name | Relevance to Howie-AI |
|---|---|---|
| NDTMS | National Drug Treatment Monitoring System | CRM data schema, reporting outputs |
| NICE QS23 | Drug use disorders quality standards | AI Counsellor clinical grounding |
| NICE CG52 | Psychosocial interventions (CBT, MI, SBNT) | Therapy model implementation |
| NICE TA114 | Opioid substitution therapy | Clinical knowledge base content |
| DSPT | Data Security and Protection Toolkit | Architecture and compliance roadmap |
| OHID | Office for Health Inequalities and Disparities | Policy alignment, From Harm to Hope |
Therapy Model Implementation
How each evidence-based therapy model is implemented in the AI Counsellor via RAG grounding.
Cognitive Behavioural Therapy
Functional analysis of triggers, cognitive restructuring prompts, coping skills training, relapse prevention planning. NICE-recommended for alcohol, cocaine, and cannabis use disorders.
Dialectical Behaviour Therapy
TIPP skill delivery (Temperature, Intense exercise, Paced breathing, Progressive muscle relaxation), distress tolerance, emotion regulation. Particularly effective for dual diagnosis (substance use + personality disorders, PTSD).
Motivational Interviewing
OARS framework (Open questions, Affirmations, Reflective listening, Summarising). Stage of change detection (Precontemplation through Maintenance) with adaptive tone. Underlies every interaction.
Eye Movement Desensitisation and Reprocessing
Psychoeducation about trauma-substance connections, EMDR preparation and safety planning. 60-80% of people with substance use disorders have trauma histories. NICE recommended for PTSD (NG116).
Acceptance and Commitment Therapy
Acceptance of cravings without acting on them, defusion from unhelpful thoughts, values clarification, committed action planning. Six core processes delivered conversationally.
Security & Data Architecture
How Howie-AI protects data and builds towards NHS compliance standards.
On-Premise & Private Cloud Deployment
Both the AI model and all data can be served from your own secure servers — fully air-gapped from the public cloud if required. No external API calls, no third-party data residency risk. Cost-effective and immune to supply-chain threats.
Server-Side Processing
All AI processing happens server-side. API keys are never exposed to the client. No PII is processed in the browser.
Encryption
Data encrypted in transit (TLS 1.3) and at rest (AES-256). All database connections use encrypted channels.
No Training on Your Data
Conversation data is never used to train any model. Where on-premise LLM deployment is used, data never leaves your infrastructure.
Audit Logging
All data access is logged with timestamps and user identifiers. Logs are immutable and retained for 90 days minimum.
Access Controls
Role-based access control. Principle of least privilege. Staff access is scoped to their operational requirements.
GDPR Compliance
Data retention policies, right to deletion, DPA with all data processors, consent management, and DPIA for high-risk processing.
Technical FAQs
How does RAG prevent AI hallucination?▾
RAG (Retrieval-Augmented Generation) grounds AI responses in a curated knowledge base — in our case, NHS-approved frameworks, NICE guidelines, and validated therapy models. The AI retrieves relevant sections before generating a response, so answers cite real content rather than generating plausible-sounding fiction.
What does DSPT compliance actually mean?▾
The NHS Data Security and Protection Toolkit is a self-assessment framework covering: Data Security Awareness training, Data Protection Officer designation, DPIA for high-risk processing, Business Continuity Plan, technical controls (encryption, access controls, audit logging). Howie-AI's architecture implements these technical controls and we're actively building towards full DSPT assessment.
How does multilingual support work technically?▾
Claude handles all 9 languages natively — no translation layer, no additional tokens. The system prompt instructs Claude to respond in the user's language. UI elements are translated via next-intl with JSON translation files for all 9 locales. RTL layout is handled via Tailwind rtl: variants on Arabic and Urdu.
What is the data sovereignty model?▾
Default deployments use the Anthropic Claude API (UK/EU data residency options available) with Neon Serverless Postgres for storage. For organisations requiring full data sovereignty, we support on-premise and private cloud deployment — the LLM and all data remain entirely within your own infrastructure, with no external API calls or third-party data exposure. No user conversation data is ever used to train any model. Data retention policies align with NHS standards (minimum 7 years for health records).
Can the AI model and data be hosted on our own servers?▾
Yes. Howie-AI supports fully on-premise and private cloud deployment for both the AI model (via self-hosted open-weight LLMs) and the data layer. This eliminates reliance on external APIs entirely — nothing leaves your infrastructure. It is also cost-effective at scale: once deployed, inference costs are fixed regardless of usage volume. This option is particularly suitable for NHS trusts, ICBs, and organisations with strict information governance requirements around data leaving their network boundary.
Can the relapse prediction model be explained?▾
Yes — explainable AI is a design principle, not an afterthought. Risk scores are accompanied by the contributing factors (engagement drop, mood deterioration, missed sessions) so keyworkers can understand and act on alerts appropriately.
Need More Technical Detail?
We can provide architecture diagrams, compliance documentation, and technical briefings for councils, commissioners, and regulated providers.
Request Technical Briefing