Problem Statement
Managing high volumes of outbound calls manually creates inconsistency, high operational cost, and limits scale. Salient needed a solution that could handle full outbound call lifecycles autonomously — from dialling and conversation to data capture and follow-up — without any manual intervention, and without limiting throughput.
Proposed Solution & Architecture
- Fully serverless voice AI platform — no infrastructure to manage, auto-scales with call volume
- Multi-turn natural language dialogue via Amazon Lex with Bedrock fallback for complex inputs
- Amazon Polly neural TTS delivering lifelike speech in multiple languages
- Automated call lifecycle: dial → converse → confirm → summarise → follow up
- Modular API architecture enabling future CRM integration without rework
- Complete call recordings, transcripts, and metadata captured for analysis
AWS Services & Technologies
What We Delivered
- Provisioned a secure, scalable cloud environment across development, testing, and production with CI/CD automation.
- Built the outbound telephony layer on Amazon Connect with intelligent call flow orchestration, caller ID management, and time-zone compliance.
- Developed multi-turn conversational flows in Amazon Lex for natural dialogue handling — intent recognition, slot filling, and escalation logic.
- Integrated Amazon Bedrock for generative AI capabilities — transcript summarisation, follow-up message generation, and handling open-ended inputs.
- Implemented Amazon Polly neural TTS for lifelike, multilingual voice output.
- Built Aurora Serverless data layer for session state, call metadata, and recording storage.
- Designed modular REST APIs for CRM integration readiness — built to connect without architectural rework.
Outcomes & Success Metrics
- Full outbound call lifecycle automated without human involvement — from dial to data capture.
- Natural, multilingual conversations delivered at scale using neural text-to-speech.
- Platform scales to thousands of concurrent calls — no performance degradation under load.
- All call data captured: transcripts, sentiment, outcomes, and follow-up actions — ready for analysis.
- CRM integration layer in place for future connectivity without rebuilding the platform.