[ AMD DEVELOPER HACKATHON: ACT II ] // UNICORN TRACK

Vibeselling
The sales agent for people who don't know how to sell

Sales tools already exist for people who know how to sell. We built the first one for people who don't — because it isn't fair that only those who already master the craft get to benefit from it. You talk about your business the way you'd talk to a friend, and the agent handles the rest.

terminal ENTER LIVE VOICE CONSOLE (DEMO_SYS) SCHEDULE PLATFORM DEMO
Optimized for sub-second latency, streaming cascade at every stage
VOXNIAC_CORE_PIPELINE_ORCHESTRATOR_v1.0.4
STREAMING E2E CASCADE
🎙️
NODE_01_ASR

Deepgram Streaming ASR (Nova-3)

Streaming WebSocket, real-time VAD events, smart formatting

VAD: real-time TURN: endpointing
⚡
NODE_02_LLM

GPT-OSS-20B (MoE)

Mixture-of-Experts, native function-calling, 131k context

ARCH: MoE CTX: 131k
🔊
NODE_03_TTS

Fireworks Native TTS

Co-located on AMD Instinct MI300X/MI350X, streaming, IPA

MODE: streaming IPA: supported
SYS_LOGS // stream_monitor.sh
[SYSTEM] Live system logs — available post-launch.
🎯

THE PROBLEM (VALIDATED FIRSTHAND)

Gabriel Bustos is CEO, chemical engineer, and certified medical interpreter. He founded SingularityOS AI LLC months ago. He has a working MVP, production infrastructure, a polished pitch deck for investors — and zero customers, because he's never sold anything in his life and doesn't know how. This isn't a hypothetical market case: it's the founder building the tool he himself needs this week, to sell to himself.

[ HOW IT WORKS ]

Three steps, no jargon

No technical forms, no configuring ICPs or sequences. Just a conversation.

🗣️

1. Tell it about your business

No technical forms — just a natural conversation. The agent asks for whatever's missing.

📋

2. The agent hands you the plan

Pricing, pitch, how to handle objections — you approve or tweak it.

📞

3. It finds leads, calls, and sells

Updates your CRM on its own. If something falls outside what it knows how to handle, it escalates to you instead of improvising with a real customer.

[ TECHNICAL_STACK ]

100% Open-Source, Served On AMD Instinct

Streaming cascade architecture: ASR → LLM MoE → TTS. No dependency on closed proprietary APIs.

🎙️

Streaming ASR

Deepgram Streaming ASR (Nova-3) — ultra-low latency, streaming WebSocket, smart formatting, real-time VAD events.

🧠

Mixture-of-Experts LLM

GPT-OSS-20B — ultra-low TTFT, native function-calling, 131k context, served on AMD Instinct.

🔊

Native TTS

Fireworks Native TTS — co-located on the same infrastructure (zero network hops), streaming, IPA pronunciation support.

🖥️

AMD Developer Cloud

App Platform — Docker container, GitHub auto-deploy, Twilio Media Streams telephony, FastAPI orchestration + state machine.

[ TECHNICAL_HONESTY ]

Why we built this

We don't promise 42ms or "10,000 concurrent streams" — that doesn't exist today. We'd rather give you a real, verifiable number than a science-fiction figure.

Our own baseline (Sharon the Hunter / Vapi) ~1,250ms
Voxniac target (E2E, streaming at every stage) ~700-900ms
Market reference (Nurix AI, live product) ~794ms

We're in the same range as an enterprise product already billing customers, not a lab ideal. 100% open-source models served on AMD Instinct hardware (MI300X/MI350X) via Fireworks AI.

[ DIRECT_CHANNEL ]

Schedule a Demo

Tell us about your business and we'll show you how the agent builds your sales plan, finds leads, and makes the calls.

HOLDING PARENT: SingularityOS AI LLC
JURISDICTION: Wyoming, USA