Voice Agents (ElevenLabs)
Human-Quality Conversational AI for Blood Shipment Logistics Notifications
Business context and structural constraints
ARC's receiving facilities process dozens of inbound shipment notifications daily. Generic IVR-style calls — with scripted prompts and keypress-based responses — created friction. Staff at receiving facilities reported ignoring or mishandling calls that felt automated, which directly affected whether facilities prepared for incoming deliveries on time. The challenge was deploying conversational AI in an environment with high variability: multiple regional centers, two delivery urgency types, different staff at the receiving end, and the requirement that every call be logged with full conversation context for compliance and QA purposes. The administrative layer needed to give ARC operations managers complete oversight without building custom tooling into each regional team's workflow.
Conversational coherence across variable medical facility contexts
Recipients at ARC-connected medical facilities show significant variation in familiarity with the notification program, background noise conditions, and communication style. Agent persona and instruction tuning went through multiple iterations based on call audit analysis before achieving consistent interaction quality across the full range of recipient behaviors.
Cross-center deduplication for shared recipient networks
Three distribution centers share overlapping recipient networks. Without explicit deduplication logic, the same facility could receive multiple calls for the same shipment from different center queues. The queue management layer implements cross-center deduplication based on recipient facility and shipment ID before any call is initiated.
The Solution
Architectural approach and implementation
Where the AWS voice system proved automation is possible, the ElevenLabs implementation proved it can sound indistinguishable from a human caller. This system replaces Amazon Polly's synthesized voice with ElevenLabs' conversational AI agents — moving from a prompted IVR-style interaction to a genuinely conversational interface where recipients can speak naturally and receive context-aware responses. The system manages outbound notification calls across three ARC regional distribution centers: Cleveland, Baltimore, and Philadelphia. When blood products are dispatched — ASAP or STAT shipments — the call queue is automatically populated, calls are initiated in priority order, and the ElevenLabs agent delivers shipment context and handles follow-up in a natural conversational exchange. Beyond the calls themselves, the platform includes a centralized administrative dashboard for queue management, call audit review, and manual order synchronization — giving ARC's operations team full visibility into the automated calling program without requiring AWS console access.
How we turned the challenge into a solution
Each stage formalizes uncertainty into a concrete engineering outcome
Audit → Dependency Map
Inventory of 17+ disparate systems, data flow mapping, identification of critical integration points and performance bottlenecks
Map → Unified Architecture
Design of event-driven microservice architecture with multi-region data residency and zero-trust security model
Architecture → Working Prototype
Document management MVP with FIDO2 authentication, AES-256 encryption, and basic workflow engine for pilot group
Prototype → Scalable Platform
Horizontal scaling to 160+ countries, multi-tenant isolation, AI document classification with 95% accuracy
Platform → Analytics Core
MyInsights recommendation engine, predictive SLA alerts, personalized delivery of regulatory updates
Core → Continuous Compliance
Automated retention policies for 160+ jurisdictions, document integrity chain, one-click audit report generation
ElevenLabs Conversational AI Integration
ElevenLabs agents are configured with ARC-specific personas and shipment context. Each call begins with a concise shipment summary — origin, destination, unit count, blood type, urgency class — and the agent handles follow-up questions, repetition requests, and transfer requests in natural, context-aware conversation. Interaction quality is evaluated continuously against call audit data.
Priority Call Queue Management
A centralized processing engine manages the outbound call queue across all three distribution centers. STAT shipments are prioritized over ASAP in the queue. Call status updates — Pending, In Progress, Success, Attempted — are written in real time and visible to administrators without requiring queue polling.
Administrative Dashboard
A TypeScript/React/Material UI web portal gives ARC operations managers full visibility into the calling program: active queue state, per-call audit history, agent performance summaries, manual order sync triggers, and queue reprocessing controls. Designed for operations managers who need oversight without technical infrastructure access.
Real-Time Order Synchronization API
RESTful APIs pull current shipment data from ARC's logistics providers before each call is initiated. The ElevenLabs agent always has current order context — unit counts, blood types, estimated delivery windows — without the risk of delivering outdated information from a stale cache.
Structured Conversation Audit
Beyond call recording, the audit system captures structured summaries of each conversation: what the agent communicated, how the recipient responded, whether key information was acknowledged, and any follow-up actions triggered. Structured audit data is accessible per-call and exportable for QA review.
The Impact
Quantitative results demonstrating the real impact of implementation on operational efficiency, infrastructure reliability, and platform scalability
Natural language delivery notifications replace IVR-style prompts — receiving facility staff engage with calls rather than routing them to voicemail
Call queue tracks every order from Pending through Success or Attempted — full status visibility for operations managers in real time
Complete call audit trail captures conversation context, agent performance, and recipient acknowledgment for compliance and QA
Multi-center synchronization keeps order data current across Cleveland, Baltimore, and Philadelphia distribution nodes
Cross-center deduplication prevents the same facility from receiving multiple calls for the same shipment from different queues
Technology Stack
Built with proven enterprise-grade technologies
Interested in a similar solution?
Discuss your project→