Case Study

Transforming Healthcare Communication with Azure and ALIANDO

Written by Alex Larsson | Dec 5, 2025 7:55:43 PM

Connecting hearts, healing with technology—powered by Azure and ALIANDO

Executive Summary

A leading communications platform serving the healthcare sector set out to revolutionise how patients, clinicians, and families connect through voice. To scale securely and introduce multilingual, AI-driven capabilities, the organisation partnered with ALIANDO to design and deliver a modern Azure architecture—bringing together real-time transcription, sentiment and tone analysis, translation, text-to-speech, and contextual search in a unified solution. The result is an extensible, cloud-native platform that supports healthcare-context communication and continuous innovation.

Client

A leading healthcare communications provider—focused on humane, multilingual voice experiences for patients and care teams.

Challenge

The platform’s vision required five advanced capabilities, all working together at scale:

  1. Transcription with automatic language detection and healthcare context.
  2. Sentiment & tone analysis across languages.
  3. Translation that preserves clinical meaning and intent in real time.
  4. Text-to-speech with natural voices across locales.
  5. Contextual search over growing message stores—without reprocessing content.

Distinct constraints compounded the challenge:

  • Real-time speech is approximately seven times costlier than batch; fast transcription options were still in preview.
  • Custom healthcare speech models demand large training datasets; without them, standard models can be similar in outcomes.
  • Tone replication at production quality requires dedicated ML R&D and was out of scope for the initial release.

Why ALIANDO

ALIANDO combined cloud architecture, LLMOps, and healthcare domain mapping to deliver a production-ready path—balancing cost, accuracy, and time-to-value. The approach included:

  • Leveraging proven Azure Cognitive Services for speech, layered with LLM-based analysis and translation for healthcare context.
  • Implementing embedding-driven search to avoid repeated analysis and enable rich, patient-specific context queries.
  • Building API-first services with CI/CD and Terraform for repeatable deployment across environments.

Solution Overview

Architecture at a Glance
  • Identity & security: Azure AD integration, Key Vault; segmented storage and patient-level security across embeddings and analysis artefacts.
  • Networking & ops: Jumpbox, VPN (P2S with certificates), Terraform IaC, Azure DevOps/GitHub Actions for CI/CD across development and production environments.
  • API Layer: Azure API Management (APIM) with four endpoints—Batch processing, Real-time processing, Real-time with TTS, and RAG Agent for retrieval-augmented generation.
  • Data & intelligence:
    • Speech: Azure Speech (Batch & Real-time) for transcription and language identification.
    • LLM: GPT-4o mini for classification, sentiment, translation.
    • Embeddings: text-embedding-3-small to power contextual search.
    • Storage: Analysis outputs and embeddings persisted (e.g., Cosmos DB/Mongo API) for fast contextual queries without reprocessing.
High-level flow: A voice message hits APIM, routes to Batch or Real-time Speech; LLM performs translation/classification/sentiment; embeddings are generated and stored; responses return via API with language, text, sentiment, and optional neural TTS output.
 
Key Capabilities Delivered
1) Transcription (Batch & Real-time)
  • Automatic language detection and healthcare-aware text output.
  • Batch for cost-efficiency; Real-time for interactive clinician-patient scenarios.
2) Sentiment & Classification
  • LLM-based multi-level sentiment and category tagging across languages.
  • Tone analysis acknowledged but deferred for a future ML track.
3) Translation
  • Bidirectional English↔supported languages; preserves clinical context and intent.
  • Real-time translation pathway for live conversation use cases.
4) Text-to-Speech (Neural)
  • Natural voices with accents and dialects; future-proof path for custom voices.
5) Contextual Search (RAG)
  • Embedding-based retrieval over patient message histories—supports text and voice queries via NLP.
  • Eliminates repeated analysis; returns contextually ranked matches.
Implementation Approach
Secure Landing Zone & Environments
  • Development and production environments with IaC, CI/CD pipelines, observability, and LLMOps integration.
API Contracts & Developer Velocity
  • Swagger definitions, C# or JS implementation, integrated telemetry and API authentication/authorisation.
Deliverables
  • Low-level designs, Terraform & pipeline setup docs, API definition & code, test results, and LLMOps setup documentation—with structured acceptance checkpoints per phase.
Phased Plan
  • Prototyping → Beta Testing → Production, with a two-month technical support window post-go-live.

Governance, Security & Compliance

  • Centralised secrets in Key Vault; role-based access via Azure AD.
  • Patient-level segmentation of embeddings and analysis data to enforce least-privilege, contextual retrieval.
  • LLMOps practices aligned to CI/CD, telemetry, and safe prompt/data handling.

Outcomes & Impact

Based on the delivered scope and architecture:

  • Faster, multilingual care communication through real-time transcription, translation, and natural TTS—reducing language barriers at the point of need.
  • Safer, more empathetic messaging with sentiment classification to flag inappropriate or negative tone for clinical review.
  • Efficient knowledge retrieval via embedding-powered contextual search, minimising repeat compute and speeding clinical workflows.
  • Operational resilience & repeatability with Terraform, CI/CD, and environment parity for continuous releases.