Skip to content
Concept

CAIA — Giving voice to medicines information

CAIA is a proof‑of‑concept AI voice assistant that turns regulator approved medicines information into clear, conversational answers, supporting patients while keeping clinicians in control.

CAIA is a concept. It does not replace healthcare professionals; it supports them. Many regulatory and legal hurdles remain, but the rollout of electronic product information (ePI) will markedly improve data quality and consistency.

Why CAIA?

Every medicines leaflet for patients contains the same line: “If you have any further questions… ask your doctor or pharmacist.” But what if you can’t? It’s late. You’re unsure. Or the question feels too minor. So you wait, or worse, you guess.

Now imagine something different. Imagine if the leaflet could talk back: a voice that answers, clearly and simply, using only trusted, regulator‑approved information. A voice that’s always there, day or night—to explain, reassure, or guide you to the help you need. That voice is CAIA.

CAIA brings medicines information to life. It transforms the patient leaflet from something you read into something you can actually talk to.

How CAIA works

  • Intent classification — identifies what the user is trying to do.
  • Grounded retrieval — for medicines topics, CAIA retrieves from a validated knowledge base (currently regulated labelling; future ePI).
  • Answer generation — responses are anchored to an authoritative source of truth to minimise inaccuracies and support auditability.
  • Multi‑turn conversation — follow‑up questions stay in context for a natural dialogue.
  • Fallback sources — non‑pharmaceutical medical questions are answered from reliable sources (e.g., NHS conditions pages).
  • Multi‑drug queries — scans every medicine mentioned to detect asymmetric interaction information.
  • Safety escalation — detects red‑flag symptoms/phrases and escalates appropriately.

Voice & languages

Automatic speech recognition models can transcribe speech in dozens of languages and auto‑detect which language is being spoken. Text‑to‑speech engines then synthesise the reply in a chosen voice and language.

Identifiers & data quality

To tame the ambiguity of drug names, CAIA can anchor searches to robust identifiers such as marketing authorisation (MA) numbers, GS1 pack barcodes, or even real‑time pack recognition via a smartphone camera.

Purpose and intended impact

  • Accessibility & availability — always on; answers when people might hesitate to contact a professional.
  • Clear, simple explanations — grounded in regulator‑approved information.
  • Empowerment through understanding — helps patients act on their medicines safely.
  • Handles complex questions — including multi‑drug interactions and asymmetric information in newer products.
  • Escalation for urgent needs — directs users to immediate care when required.
  • Linguistic versatility — multilingual transcription and synthesis.

High‑level flow (concept)

  1. User speaks ➝ automatic speech recognition (ASR) ➝ text.
  2. Intent classification ➝ choose medicines / general health / other.
  3. If medicines ➝ retrieval from validated KB (regulated labelling; future ePI) ➝ generate grounded answer with citations.
  4. Check for safety triggers ➝ escalate if necessary.
  5. Text‑to‑speech (TTS) ➝ spoken response; keep context for follow‑ups.

We are actively testing these components and exploring how AI can drive safer, smarter prescribing.


Frequently asked questions

What is CAIA?

CAIA is a proof-of-concept AI voice assistant that turns regulator-approved medicines information into clear, conversational answers. It supports patients and clinicians by retrieving from validated sources, generating concise explanations with citations, and maintaining context across follow-up questions. CAIA does not replace professional judgement.

Where does CAIA get its information?

CAIA is grounded in authoritative sources such as regulator-approved labelling and future ePI. For out-of-scope general health topics it signposts reputable public guidance. Answers include links or citations back to the specific section used, supporting transparency and auditability.

How does CAIA handle safety and urgent issues?

CAIA detects red-flag symptoms and risk phrases. For anything urgent, ambiguous or beyond scope, it defers to a healthcare professional or emergency services as appropriate. Its role is supportive and informational; clinical decisions remain with qualified professionals.

Does CAIA keep a record of conversations?

In concept form, CAIA retains minimal data for quality and safety evaluation. Any deployment would follow strict governance: clear notices, opt-in where required, limited retention, and secure handling aligned with applicable regulations. CAIA does not create patient records or replace clinical documentation.

What languages and voices does CAIA support?

CAIA can use automatic speech recognition to transcribe many languages and detect the spoken language. Text-to-speech then synthesises the reply in a chosen voice and language. Voice selection, pronunciation and accessibility are part of ongoing tuning in the concept roadmap.

What’s next for CAIA?

The roadmap explores structured ePI ingestion, improved guardrails and audit trails, and accessibility enhancements for voice UX. We are evaluating pilots with medicines information teams to test safety, usefulness and governance in real-world workflows.


Interested in partnering on CAIA?

We’re looking for collaborators across medicines information, pharmacovigilance, and digital health. If you’d like to evaluate feasibility, co‑design use cases, or explore pilots, please get in touch.

Get in touch

Follow us on LinkedIn or email steve@pharmadvisor.org.uk.