Customer Story
5 min read
read
How Relay Hawk Launched an AI Voice Studio in Record Time
See how Relay Hawk, a solo-founder startup, built a production AI voice agent studio on SignalWire and launched in just months.
Dani Plicka
Content Marketing Manager
Subscribe
Tags
Voice AI
AI Agents
PUC
Contact Center
Latency & Performance
Relay Hawk: AI Voice Agent Studio
Solo founder Justin Massey built Relay Hawk, an AI voice agent studio for answering services and call centers, on SignalWire’s Programmable Voice API and AI tooling. Replacing rigid call-flow diagrams with natural-language prompts let him take the platform from prototype to production in months, not years.
Overview
Relay Hawk helps answering services and call centers rapidly deploy AI phone bots tailored to specific customer needs. After hitting roadblocks with legacy LLM-based workflows, CEO Justin Massey turned to SignalWire Voice AI.
A solo founder working by himself, Massey built a production-grade voice agent studio in 5 months. This studio integrates with call centers and handles transfers and escalations to humans reliably.
With SignalWire AI, Massey moved away from rigid flowcharts in favor of a flexible, prompt-based development model This shift to plain-text prompts drastically reduced development time so he could focus on customer outcomes instead of infrastructure.
About Relay Hawk
Relay Hawk modernizes call centers’ tech stack with conversational AI and agentic workflows. With an AI IVR overlay that integrates with existing systems, call centers can deploy low-latency AI phone agents that triage routine calls and escalate complex cases to human agents without replacing any of their core infrastructure.
Massey didn’t initially set out to build a product. His original plan was to educate mid-sized contact center owners on how to adopt AI before AI-native competitors left them behind. But as he explored what it would take to deliver production-quality voice agents himself, testing solutions like Twilio and OpenAI, it became clear that stitching together the voice AI stack with low enough latency for lifelike conversations was nearly impossible.
“We were able to accomplish what our customers needed a lot more quickly with SignalWire AI than with other standalone call flow systems.”
Justin Massey
Founder and CEO, Relay Hawk
The Challenge
Relay Hawk's target customers were mid-sized contact centers that handled inbound and outbound calls on behalf of other companies and government organizations. These contact centers were losing customers to AI-first offerings. Massey planned to create an online course, teaching them to build their own AI voice agents.
In his learning process, Massey initially relied on standard LLMs, integrating with OpenAI to infer possible caller paths and journeys in call flows. But this approach introduced several pain points:
Flows were too strict, requiring tight definitions of every possible user path.
Iteration was slow, with every change needing diagram and integration updates.
Telephony overhead like handling WebSockets, SIP, and latency became a bottleneck.
There was no easy way to scale conversational complexity or respond dynamically to callers.
The Solution
After testing SignalWire's Programmable Voice API and discovering how much lower the latency was, Massey realized the bigger opportunity: build a new product. Not just a single AI voice agent, but a platform for building multiple voice agents. These are sometimes known as “agent studios” or AI IVRs. With SignalWire, invoking the entire voice AI pipeline was a few lines of code. Phone bot behavior could be defined using natural language prompts and easy-to-read markup, without having to define every possible path of the call flow.
By replacing rigid workflows with natural language prompts and letting SignalWire handle the telephony stack, Relay Hawk was able to ship faster and scale. Prompt-based control over call flows makes changes fast and flexible, while there is no need to manage infrastructure. SignalWire handles SIP and audio quality, delivering a low latency end result.
The Results
Faster delivery and lower latency | Focus on the caller | Telephony done for you | Faster market response |
|---|---|---|---|
Bots that used to take days to design and ship now go live in hours, keeping Relay Hawk’s customers ahead of schedule. | With infrastructure abstracted away, Massey focuses entirely on improving caller experience and AI interaction quality. | SignalWire’s Programmable Unified Communications (PUC) model handles routing, SIP, PSTN, and audio latency, so Massey doesn’t have to build or maintain that layer himself. | Because SignalWire handles the infrastructure, Massey can adapt his roadmap to customer feedback and a fast-moving voice AI market in hours or days rather than weeks or months. |
“I don't have any desire to deal with the underlying telephony if it's been solved in a good way. Where the calls are low-latency, I'm willing to pay a price for that. It saves me time and allows me to ship faster.”
Justin Massey
Founder and CEO, Relay Hawk
How long does it take to build an AI voice agent?
Timelines vary widely by approach. Teams stitching together separate LLM, telephony, and speech components from scratch often spend many months on integration work alone. Teams building on a unified voice AI platform that handles telephony, latency, and orchestration can go from prototype to production in weeks to a few months, even as a solo developer.
Why does latency matter so much in AI voice agents?
Delays of even a few hundred milliseconds between when a caller finishes speaking and the agent responds break the illusion of natural conversation and make bots feel robotic. Low latency depends on the underlying telephony and speech infrastructure — not just the language model — so the choice of voice platform has a direct effect on how "human" an agent feels.
What infrastructure do you need to build a production AI phone agent?
At minimum: a way to receive and route calls (SIP/PSTN handling), real-time speech-to-text and text-to-speech, an LLM or reasoning layer to drive the conversation, and a mechanism to escalate to a human when needed. Historically, teams assembled these from separate vendors; newer platforms increasingly bundle them so a single API call invokes the whole pipeline.
How does an AI voice agent hand off a call to a human?
Most production systems support an "escalation" or "transfer" trigger — the agent recognizes it's hit the edge of what it can resolve and hands the conversation to a live agent, ideally passing along context (what the caller said, what's been tried) so the caller doesn't have to repeat themselves.
Can one person build a production-grade AI voice agent platform?
Yes, increasingly so — the bottleneck used to be assembling and maintaining low-latency telephony infrastructure, which typically required a team. Platforms that abstract away SIP, PSTN, and audio-quality handling let a solo developer focus on conversation design and ship a full product without hiring a backend team.
Related Resources
