// the find
neural-maze/realtime-phone-agents-course
Build realtime AI voice agents using FastRTC for low-latency streaming, Superlinked for vector search, Twilio for live phone calls, and Runpod for scalable GPU deployment.
A multi-week course (not a library) walking through building a phone-based AI voice agent for a fictional real-estate call center: FastRTC for the realtime audio loop, Twilio for inbound/outbound calls, Superlinked for multi-attribute property search, and RunPod-hosted STT/TTS models. It's aimed at ML/AI engineers who want to see a full voice-agent stack wired together end to end, not someone looking for a drop-in library.
Actually covers the whole pipeline instead of stopping at a demo: STT, TTS, hybrid retrieval, real Twilio telephony, and Opik tracing with per-method instrumentation and prompt versioning are all wired up and runnable. It gives you multiple swappable STT/TTS backends (local Moonshine/Kokoro, Groq/TogetherAI APIs, self-hosted RunPod Faster Whisper/Orpheus) so you can actually compare latency and quality instead of being stuck with one choice. The Superlinked lesson is a genuinely useful worked example of combining text, number, and categorical spaces into one search index with runtime-adjustable weights, which most retrieval tutorials skip.
It's a tutorial artifact on a release schedule — a meaningful chunk of the explanation lives in gated Substack articles rather than the repo itself, so the code alone is only half the story. Getting it running costs real money and setup time: Twilio, RunPod GPU pods, Qdrant Cloud, Opik, and Groq/TogetherAI keys are all required before you see anything work, which is a lot of friction for something billed as a learning project. The property search demo runs on a single small CSV, so it won't teach you anything about retrieval behavior at real scale. There's no test suite anywhere in the tree — the only CI is Docker build/push for the RunPod images — so despite the 'production-ready' framing, nothing actually validates the agent pipeline itself.