Skip to content
← work

Case study

iCall — Real-Time Voice AI Communication Platform

Sr. Full Stack Developer · Freelance · 2025

An enterprise real-time Voice AI communication platform utilizing Asterisk PBX, AudioSocket, Pinecone vector search, and Pulumi IaC.

▸ Sub-second response times on live, bidirectional voice conversations

▸ RAG-driven knowledge retrieval executed mid-call for context-aware agent responses

▸ Infrastructure automated across AWS/Azure with Pulumi for dynamic scaling and high availability

Problem

Holding a natural, low-latency voice conversation with an AI agent over real telephony means solving live bidirectional audio transport, sub-second model turnaround, and pulling accurate context mid-call — all at once.

Approach

Architected the system on Asterisk PBX with SIP and VoIP for bidirectional voice over internet telephony, and built a high-throughput AudioSocket pipeline streaming live raw audio into a hybrid Python (FastAPI) and Azure OpenAI processing framework. Pinecone provided contextual RAG retrieval executed mid-call, and Pulumi (IaC) provisioned and automated the cloud footprint across AWS and Azure.