These 7 voice AI projects just blew us away

Over the past few weeks, developers from around the world joined the AssemblyAI Voice Agents Challenge on DEV.to to build with our most advanced real-time transcription API: Universal-Streaming. With ultra-fast, sub-300ms latency and superhuman accuracy, Universal-Streaming unlocks real-time voice experiences for everything from live gaming to customer support automation.

Curious about what the challenge involved? Check out the original brief and rules here.

We were blown away by the creativity and technical skill of the submissions. Over 60 developers submitted real-time voice agents, from magical games to inclusive AI assistants. After careful review, we're thrilled to announce our top 3 winners along with 4 other incredible finalists who made the shortlist.

Winners of the AssemblyAI Streaming Voice Agent Challenge

1. Build an application where sub-300ms latency matters

Real-Time Performance Winner: Hogwarts Spell Caster – Real-Time Voice Magic by @axrisi

‍

Wave your wand! This voice-controlled Harry Potter game lets users cast spells using natural voice commands, transformed instantly into keyboard actions. A magical use case where latency really mattered.

GitHub Repo to build A real-time voice-controlled spell casting system

2. Build a specialized voice agent with deep domain knowledge that can learn and improve from conversations

Domain Expert Winner: AI✧Debate by @nadinev

‍

The AI Debate Room sharpens your arguments by providing an AI Opponent that delivers real-time, topic-aware, and domain-expert counter-arguments. It achieves this by retrieving relevant content, assembling it into context, and utilizing a Large Language Model (LLM) to craft concise rebuttals from an opposing viewpoint.

GitHub Repo to build a real-time AI debate platform

3. Build a voice agent that automates real business processes 

Business Automation Winner: Wynnie 🦄 — 𝘠𝘰𝘶𝘳 𝘚𝘩𝘰𝘱𝘱𝘪𝘯𝘨, 𝘯𝘰𝘸 𝘰𝘯 𝘈𝘶𝘵𝘰𝘱𝘪𝘭𝘰𝘵! by @neilblaze & @achalbajpai

Wynnie 🦄 is an autonomous voice-powered shopping agent that makes online shopping effortless and accessible. Built using AssemblyAI’s real-time speech recognition with automatic language detection, Wynnie takes care of everything from finding deals to managing checkout. Designed with inclusivity in mind, it serves both digital natives and underserved groups—like elderly users—who are often left behind by today’s interfaces. With multi-agent orchestration, real-time product search, smart deal hunting, and a clean PWA experience, Wynnie reimagines the future of voice-first e-commerce.

GitHub Repo to build a smart autonomous AI shopping agent

Real-Time STT Challenge Shortlist

These projects stood out for their innovation, usability, and thoughtful implementation of voice agents powered by AssemblyAI:

Accessibility + Inclusion Agent:

• Voice of Voiceless – Enabling the Voiceless to Communicate by @mohamednizzad
  Accessibility-focused voice agent that helps users with speech impairments communicate in real time using transcription and synthesized responses.

Education + Coaching Agent:

• Grasp, Articulate & Refine by @divyasinghdev
  A voice-based academic coach to help with studying, summarizing, and recalling key concepts. While this project doesn’t use the Universal-Streaming endpoint, its creativity and potential impact earned it a spot on our shortlist

Healthcare & Appointment Setting Agent:

• CareSetu AI by @saket_jha_a89aca5daba5e8c
  A healthcare voice agent for appointment scheduling and basic health queries.

Voice Assistant for Daily Tasks:

• Elder Care Companion by @vivek-aws
  A voice assistant designed to support aging adults with daily tasks and companionship.

What Can You Build with Universal-Streaming?

The Voice Agents Challenge showcased just a glimpse of what's possible with AssemblyAI’s Universal-Streaming API. From accessibility apps and real-time education tools to interactive gaming and healthcare support agents, developers proved that sub-300ms latency and state-of-the-art speech-to-text can enable truly responsive voice-driven applications.

This is just the beginning. If you're interested in building with Universal-Streaming, sign up for your free API key (no credit card required) and get $50 in credits to start building instantly.

Building a voice product at your company? AssemblyAI's Universal-Streaming is production-ready and scales with you.

Check out our cookbook to see workflows and examples of AssemblyAI’s streaming speech-to-text model. 

Want to see what others are building — or share something of your own?
Check out the full Voice Agents Showcase to explore every submission from the challenge.

If you’ve built something cool with AssemblyAI, post it on X using #builtwithassemblyai — we’re always looking for great projects to feature in our blog, newsletter, and beyond.

‍