Building an AI-Powered Patient Interview Platform on AWS

Overview

A clinical research organization specializing in cognitive health approached MakBPO with an ambitious vision: replace slow, manual, staff-intensive patient interviews with an automated platform that could conduct structured cognitive assessments by phone – at scale, consistently, and in full alignment with healthcare data-protection standards.

MakBPO took the concept from idea to a fully architected, working platform: a serverless AWS system that registers patients, schedules and places automated calls, runs 15–20 minute AI-guided cognitive interviews, records and transcribes them, and feeds every interaction into an analytics-ready data warehouse.

The engagement ran across roughly seven months – two months of discovery and architecture, followed by four to five months of development – delivered by a focused four-person team led by a dedicated project manager.

The Challenge

Cognitive-health research depends on consistent, high-quality patient interviews. Conducted manually, these interviews are expensive, hard to scale, and difficult to standardize: clinical staff must schedule each call, conduct it live, take notes, and transcribe findings, all while juggling availability across patients and time zones. Quality varies between interviewers, geographic reach is limited by staffing, and the administrative overhead grows linearly with every new patient.

The organization needed a platform that could:

• Conduct standardized cognitive interviews automatically, without a clinician on every call
• Schedule and manage hundreds of calls with intelligent retry handling
• Capture, store, and transcribe interviews securely
• Enforce strict data isolation across multiple research sites
• Meet the privacy and security expectations of regulated healthcare data
• Scale from a pilot to high call volumes without re-architecture

This was not a simple CRUD application. It required orchestrating telephony, an AI conversation engine, speech-to-text transcription, secure storage, and analytics – all stitched together into a reliable, compliant whole.

The Solution

MakBPO designed and built an end-to-end, event-driven platform on AWS. The system spans a secure provider portal, a serverless API and business-logic layer, an operational datastore, a voice and transcription pipeline, and a dedicated analytics warehouse.

Provider portal

A clean, role-aware web portal lets site managers register patients, schedule calls, upload assessment and consent documents, monitor call status on a live dashboard, and submit change requests for administrative review. Patients are managed through a dual-identifier system – a human-friendly public ID for day-to-day use (e.g. ABC-001234) and an internal UUID for all system operations – keeping front-of-house interactions simple while internal references stay robust.

Automated interview pipeline

When a patient is registered, the platform schedules their interview and, at the appointed time, triggers a serverless call-initiation flow. It assembles the patient’s context, places the call through an enterprise telephony provider, and hands the live conversation to an AI voice agent that conducts a structured 15–20 minute cognitive interview – covering daily-routine questions, hobbies and interests, and narrative-recall assessment tasks. Calls are recorded, transcribed through an AI speech-to-text engine, and the structured results are stored automatically.

Intelligent retry logic

Real-world calling is messy – voicemail, busy signals, no-answers, declines. MakBPO implemented a four-stage retry strategy (immediate, +5 minutes, +1 hour, +24 hours) with per-scenario handling for busy signals, technical errors, and invalid numbers, plus automatic site notification when attempts are exhausted. This turns an unreliable channel into a dependable one.

Consent and SMS compliance

The platform handles explicit SMS opt-in at registration, sends confirmation and reminder messages, and honors STOP/opt-out requests automatically – capturing consent timestamps for a clean compliance trail.

Analytics warehouse

Operational data lives in a fast NoSQL store optimized for real-time call handling. A nightly ETL process transforms and loads that data into a PostgreSQL star-schema warehouse – dimension and fact tables for patients, sites, providers, calls, interviews, documents, and change requests, plus pre-aggregated daily metrics – giving the research team a clean, query-ready foundation for analysis and reporting.

Architecture Highlights

MakBPO built the platform fully serverless and event-driven, so the system scales with demand and the client pays only for what it uses.

• Event-driven orchestration: using scheduled triggers and queue-based processing, so calls, transcriptions, and data loads happen reliably and asynchronously.
• Serverless compute: for the API and for each discrete job – call initiation, webhook processing, transcription, retry scheduling, and ETL – each independently scalable and isolated.
• Purpose-fit data layer: a low-latency NoSQL operational store for live call handling, paired with a relational analytics warehouse for reporting – the right tool for each job rather than one compromise database.
• Secure media pipeline: encrypted audio storage with lifecycle policies, time-limited signed access URLs, and an automated transcription queue with dead-letter handling for resilience.
• Strict multi-site isolation: every query is automatically scoped to the requesting user’s site, enforcing that research sites can never see one another’s data.

The platform was engineered to meet demanding performance targets: sub-500ms API responses, 1,000+ requests per minute, 50+ concurrent live calls, and a 99.9% uptime design goal – with auto-scaling and high-availability patterns built in from the start.

Security & Compliance

Because the platform handles sensitive patient health information, security and compliance were designed in from day one rather than bolted on:

• Encryption at rest (AES-256 / KMS) and TLS in transit across all services
• Role-based access control with site-level data isolation
• Comprehensive audit logging and trail retention aligned to healthcare retention requirements
• Soft-delete and versioning for documents, preserving an immutable history
• Secrets management with automated credential rotation
• Web application firewall, rate limiting, and signature validation on external webhooks

What MakBPO Delivered

• A secure, role-aware provider portal (login, patient registration, dashboard, document upload, and change-request workflows)
• A complete serverless API and business-logic layer covering authentication, patient management, documents, requests, and webhooks
• An automated voice-interview pipeline integrating telephony, an AI conversation engine, and AI transcription
• Intelligent multi-stage retry and call-concurrency management
• SMS consent, confirmation, reminder, and opt-out handling
• A fully specified operational data model and a PostgreSQL analytics warehouse with a nightly ETL pipeline
• Complete infrastructure design across compute, storage, networking, security, and monitoring
• Detailed technical documentation: API specifications, infrastructure spec, business-logic rules, data models, and ETL mapping

Technology Stack

• Cloud & infrastructure: AWS (serverless), API Gateway, Lambda, ECS Fargate, EventBridge, SQS/SNS, CloudFront, CloudWatch, CloudTrail, KMS, Secrets Manager
• Data: DynamoDB (operational), Amazon RDS for PostgreSQL (analytics warehouse), Amazon S3 (encrypted media & documents)
• Voice & AI: Telephony platform integration, AI voice conversation engine, AI speech-to-text transcription
• Application: REST API backend, modern web front-end provider portal
• Practices: HIPAA-conscious design, infrastructure-as-code patterns, event-driven architecture, star-schema data warehousing

The Team

MakBPO delivered the project with a lean, senior team: a dedicated project manager leading the engagement, supported by a DevOps engineer, a backend developer, and a frontend developer – proving that a focused, well-coordinated team can deliver genuinely sophisticated, regulated-industry software.