# OpenAI Voice Hack Night: Project Gallery

- **Event:** [OpenAI Voice Hack Night](https://cerebralvalley.ai/e/openai-voice-hack-night)
- **When:** Wed, May 27 at 3:00 – 9:00 PM (PDT)
- **Where:** 1515 3rd Street, San Francisco, CA
- **Hosts:** [OpenAI](https://cerebralvalley.ai/u/openai), [Cerebral Valley](https://cerebralvalley.ai/u/cv)
- **Projects:** 59 (4 placed)
- **Page:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery

## Projects

### 1. Coval

Integrated the GPT Realtime 2 model as one of the out of the box Agent options so people can test their system prompts in a variety of cases! Then they can use a bunch of different metrics to evaluate the performance of that agent. Some of these include LLM Judge metrics as well. Notice how good the latency is for the GPT Realtime 2 model. Also make note on how the interruption rate increases when hmmm and ahhh are scattered throughout the conversation.

- **Team:** [Kobi Hudson](https://cerebralvalley.ai/u/KobiHudson)
- **Demo video:** https://youtu.be/GQm5ZmyJe2Q
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=1

### 2. Athena (AI Voice Coach for Thinking, Speaking, and Pitching)

We as humans struggle to practice a technical presentation or product pitch by ourselves or find a human who listens carefully and gives actionable feedback. We built a real-time AI voice coach using Codex and the OpenAI Realtime API that listens during live practice sessions and provides feedback in voice or text to help identify gaps and improve your craft. The GPT model can respond when explicitly asked or proactively detect hesitation, pauses, or confusion in your voice. Potential use cases include product pitches, complex technical presentations, and sales calls.

- **Team:** [Kazem Jahanbakhsh](https://cerebralvalley.ai/u/jahan)
- **Demo video:** https://www.youtube.com/watch?v=9SxY4EAMBOs
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=2

### 3. StoryCast

StoryCast is a voice agent for global storytelling. A user can speak a short story in one language, and StoryCast turns it into a multilingual illustrated audiobook. It transcribes and punctuates the spoken story, extracts a story knowledge graph of characters, goals, conflicts, actions, and outcomes, translates the story into multiple languages, narrates it back, and generates scene illustrations.

The problem it solves: stories, lessons, family memories, and creator content often get trapped in one language or format. StoryCast helps creators, parents, educators, audiobook makers, and language learners turn spoken stories into accessible multilingual content.

- **Team:** [Isha S](https://cerebralvalley.ai/u/ulat)
- **Demo video:** https://youtu.be/PHMHC13AJIw OR https://youtu.be/LABfnZIkzK0
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=3

### 4. Enata

Automation and intelligence for field sales, powered by the realtime models.

- **Team:** [Marty Moesta](https://cerebralvalley.ai/u/mamoesta), [Smit Shah](https://cerebralvalley.ai/u/shahsmit), [Justin Bedard](https://cerebralvalley.ai/u/Enata)
- **Demo video:** https://www.youtube.com/watch?v=4YR8rmBHX0U
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=4

### 5. Sign Bridge

SignBridge is a two-way communication bridge between sign-language users and the non-signing people around them, built for moments where reading or typing on a phone isn't an option - for Deaf users who are also blind or low-vision, for children, and for elderly signers. Using only a webcam, SignBridge recognizes a short signed phrase, rewrites it into a natural spoken sentence, plays it out loud in the listener's language, and renders a clear pictogram card so the request can be heard and seen at the same time. In the reverse direction, the hearing person's text reply is simplified into accessible language and broken into a short signer-facing storyboard of visual beats. The whole loop is powered by a chain of OpenAI products - Responses API with vision for sign recognition, text models for translation and simplification, the Audio Speech API for voice, and Image Generation for the communication card — wired together in a Next.js app. This hackathon build ships as a web demo for portability and quick iteration, but the experience is designed for a native mobile app, where always-on camera access, one-tap launch from a lock screen, offline-friendly phrase packs, and on-device haptics would make it genuinely usable in the street, the train station, or the store aisle.

- **Team:** [Jiyun Kim](https://cerebralvalley.ai/u/jiyun)
- **Demo video:** https://drive.google.com/file/d/1IpUjdNE2sBMmLvNz3ypvIWW3HNF46Sl5/view?usp=sharing
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=5

### 6. Voice for mapping-ai

Voice commands that help navigate and search on mapping-ai.org.  The OpenAI tools used were Codex and gpt-realtime-2

- **Team:** [Daniel Pang](https://cerebralvalley.ai/u/danp)
- **Demo video:** https://youtu.be/c1yIOBxSIgU
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=6

### 7. gopal

artificial goblin intelligence.

gopal is Her, but a goblin.

a spatial AI companion that lives in your view (Apple Vision Pro, any video stream) as an animated character. uses GPT Realtime 2 for natural voice conversation with sub-second latency, accepts live camera input so it can see and react to your surroundings, and renders with idle, dance, and reaction animations. plays video games with you, talks banter, understands spatial intelligence. a chaotic goblin that sees what you see, talks back in real time, and actually knows where you are via a persistent knowledge graph that gives it long-term memory and context.

check it out live! 
goblin overlay: gopal.stephenhung.me

please let us demo live trust this is agi

- **Team:** [Joshua Lin](https://cerebralvalley.ai/u/qtzx06), [Stephen Huaien Hung](https://cerebralvalley.ai/u/stpnhh), [Matthew Kim](https://cerebralvalley.ai/u/matthewykim)
- **Demo video:** https://www.youtube.com/watch?v=rxYpBmyHk1g
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=7

### 8. Autoprompter

Any content creator or public speaking platform needs a lot of practice, that why teleprompter came up.

The project enhances a teleprompter to use your Audio/speech to guide you into the right context, based on the audience feedback, live responses, your voice tone etc

- **Team:** [Div Agarwal](https://cerebralvalley.ai/u/div_vi)
- **Demo video:** https://drive.google.com/drive/folders/1Q7rWddzYeMk6fobOzSQzJeGAzw0yz7Gb
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=8

### 9. Paradigm outreach

Parry is an agentic AI SDR that runs your entire outbound operation through a voice interface. You call him in your browser, tell him what you want, and he researches leads, writes custom emails, manages deliverability, handles replies, and books meetings. A single request triggers 21 tool calls before handing off to Generative UI, which renders live cards, pipeline stats, lead detail views, and strategy slides in sync with his voice. He has cross-channel memory across voice and SMS, and a custom voice prompt that morphs per customer at onboarding.

- **Team:** [Manuel David](https://cerebralvalley.ai/u/ParadigmCEO)
- **Demo video:** https://youtu.be/tEa7TncW2fk
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=9

### 10. Room

I spend all day jumping between 5-10 agent threads across Codex app and iterm tabs. I set agents on tasks and let them work autonomously, checking in on their status and guiding next steps when they are complete. Executives do this with teams and they do it by speaking with managers representing workflows. Room lets you be the executive in the room- agent managers report status updates, you can patch in managers for specific workflows you want to check on, and you can instruct next steps - all from one place, all with your voice. A central room where decisions are made and results are reviewed. Kick your feet up on your mahogany desk, settle back into your large leather chair, and start telling people what to do. You are the boss in this Room. 

All made today during the hackathon, GH - https://github.com/callumreid/room

- **Team:** [Callum Reid](https://cerebralvalley.ai/u/callum)
- **Demo video:** https://www.loom.com/share/bdcb3c4941b849d1b9b5a8898d0e5593
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=10

### 11. Woodstock

SnoopyAI is a voice agent platform for dry cleaning businesses to handle inbound customer calls (e.g. requests to expedite dry cleaning) or outbound calls (e.g. pickup reminders).

- **Team:** [Aayush Gandhi](https://cerebralvalley.ai/u/aayushgandhi), [Dhruva Vutukury](https://cerebralvalley.ai/u/dhruvavutukury)
- **Demo video:** https://docs.google.com/document/d/1FwSZj7Zv2C7uD3FzpEITvfJFsjdQrU9sxu-YJw3mb4g/edit?usp=sharing
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=11

### 12. (hooked on) Phonex

My 3-year old has shown interest in learning how to read, but all of the phonics-based apps we've tried are cumbersome, rigid, and require an adult to use effectively. 
  
So I built Phonex, an app to help 3-7 year-olds learn how to read. 
   
It was build for the hackathon entirely with Codex and uses a combination of OpenAI's Realtime API, speech-to-text (for "grading" phonemes), and text-to-speech (for generating phonemes/words), plus a formal state machine to control and track progress. 
  
Github repo: https://github.com/jumploops/phonex

- **Team:** [Adam Williams](https://cerebralvalley.ai/u/adamloops)
- **Demo video:** https://youtu.be/D20hkC9yHnM
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=12

### 13. Walkie Talkie

**Walkie-Talkie** is a real-time thinking partner orchestrating a dual model setup of `gpt-realtime-2` and Codex. It pairs a live voice interaction model with parallel background agents that handle tool calling, code execution, and generative UI, so the conversation never stops while work happens behind the scenes.


### Why users need this

Today, using an AI agent means watching it work. You prompt, you wait, you read, you prompt again. Voice doesn't fix this if it's still turn-based. What people actually want is to think out loud while things get done. Walkie-Talkie lets you talk through a problem, ask for a chart, pivot to a different question, request a code change, all in one continuous conversation, while each task runs in the background and results surface as they complete.


### The architecture

When Thinking Machines released their interaction model, I recognized the same architecture I'd been building independently. Their thesis: for interactivity to scale with intelligence, it must be part of the model itself. I don't fully agree. Model plus harness becomes a powerful, collaborative agent. These are two means to the same end. Interactivity is a type of user experience, and as long as users get that experience, it doesn't matter whether the system uses a full-duplex model or cascaded scaffolding. Walkie-Talkie proves this with OpenAI's realtime voice stack. `gpt-realtime-2` handles the hard interaction problems (sub-230ms responses, pause detection, mid-sentence pivots, simultaneous listening and speaking). Background agents handle the heavy work.


### Parallel agents and context management

The background is not a fixed set of tool functions. Each background agent is a coding agent with shell access in a persistent sandbox, the same pattern that makes CLI-based agents like Codex so powerful. Instead of predefining every tool (generate_chart, run_query, etc.), the agent gets intent from the voice model and figures out *how* autonomously. Multiple agents run simultaneously, each handling a different task. The front-end coordinator dispatches tasks, tracks which agents are working, and manages the context bridge between the voice conversation and each background agent. When an agent finishes (a chart, a file, a generative UI artifact), the result routes back to the voice model, which weaves it into the conversation naturally. This makes the system an agent orchestration layer with voice as the interface.


### Why this is the correct architecture

I analyzed the five capabilities needed for natural voice interaction: speaking during user speech, pause-vs-endpoint detection, real-time semantic processing, micro-responses, and simultaneous input/output. These capabilities conflict with the demands of heavy tool execution. Separating them isn't a shortcut. It's the correct design approach.


### Links

- **Using:** gpt-realtime-2 (OpenAI Realtime API), Codex (CLI/SDK)
- **Research:** https://lilyzhng.github.io/posts/interaction-model/
- **Demo:** https://lily-walkie-talkie.vercel.app/
- **GitHub:** https://github.com/lilyzhng/walkie-talkie

- **Team:** [Lily Zhang](https://cerebralvalley.ai/u/lilyzhng)
- **Demo video:** https://lily-walkie-talkie.vercel.app/ https://www.youtube.com/watch?v=6jWTQ3CaYrg
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=13

### 14. Surgical Triage

I am a practicing hand and microsurgeon, and one of my specialties is reattaching amputated digits. A major part of that work is receiving urgent transfer calls from emergency departments hours away, often while I am already operating. These calls matter, but they are frequently incomplete: missing photos, x-rays, ischemia time, amputated-part handling details, or basic transfer logistics. They can distract from the operation while still leaving me without the information I need to make a fast decision.

I built this product to turn that chaotic transfer call into a realtime coordinated workflow. The system answers the transfer line, speaks with the referring ER doctor, reviews uploaded photos and x-rays during the call, identifies missing or inadequate imaging, checks surgeon-specific hand-surgery criteria from a library of clinical skill files, corrects urgent errors like improper amputated-part preservation, and builds a live transfer packet for the receiving surgeon as the conversation unfolds.

Once the patient is accepted for transfer, the agent can use the same structured information to coordinate next steps, including calling the operating room to help schedule the surgery. The goal is to remove the repetitive intake burden, protect the operation already in progress, and make the final handoff faster, safer, and more focused.

- **Placement:** 3rd Place
- **Team:** [Brian Pridgen](https://cerebralvalley.ai/u/brian_md)
- **Demo video:** https://youtu.be/Sa-mFTEhV1U
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=14

### 15. Halda

We are leveraging the realtime 2 speech to speech model and its tool-calling capabilities to augment the form building process. Rather than manually building and configuring a form within our product-s form builder, the voice assistant can make tool calls to add, delete, and modify any components of the form. This includes image generation and styling, as well as basic input types and question formats.

- **Team:** [Jonny Linford](https://cerebralvalley.ai/u/jonnylinford), [Riley Ross](https://cerebralvalley.ai/u/riley13ross), [Benjamin Dazey](https://cerebralvalley.ai/u/bdazey), [Joshua Richardson](https://cerebralvalley.ai/u/joshuarichardson)
- **Demo video:** https://drive.google.com/file/d/19ZSso-Vmtx3M07OfTFECW6lZx9EdNBP-/view?usp=drive_link
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=15

### 16. OWEN

A voice agent to answer your questions about github repos.

- **Team:** [Tommy Purcell](https://cerebralvalley.ai/u/tommypurcell)
- **Demo video:** https://www.loom.com/share/83f08daa8c7a4540af8bcf972290e8a2
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=16

### 17. Alva Health

Nobody wants to wait on hold for minutes or hours — especially doctors who bill by the hour. Our AI agent waits on the line for you, handles the call, and brings a human into the process only when needed.

- **Team:** [Ruben Sandoval](https://cerebralvalley.ai/u/bennsandoval)
- **Demo video:** https://youtu.be/XGIOR25l9zo
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=17

### 18. Nekko　Real Time Translate

Nekko Live Interpreter is an iOS app that turns your iPhone into a real-time bilingual voice interpreter, designed specifically for hackathon demos and pitches where Japanese presenters need to speak to an English-speaking audience (and vice versa).

You hold a single big button, speak in Japanese (or English), and the moment you release, a cute cat character (Nekko) speaks back the translation in the other language using OpenAI's Realtime API — all voice-to-voice, no typing, no waiting.

The whole UI is just "Nekko + one button" on purpose: during a demo or Q&A you don't have time to look at transcripts or pick a direction. Push-to-talk solves the noisy-room problem (we kept getting interrupted by ambient conversations), and the strict translator-only prompt prevents the model from replying conversationally to phrases like "translate this".

Built originally as a Mistral-based meeting recorder, we rebuilt the interpreter flow on top of OpenAI's gpt-realtime in ~1 day for this hackathon.

- **Team:** [Shohei Yukawa](https://cerebralvalley.ai/u/Shohei)
- **Demo video:** https://youtube.com/shorts/9q6bZZY6b-M?feature=share
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=18

### 19. Language Stream

Use OpenAI's voice real time models to live translate livestream Twitch audio.

- **Team:** [Petros Hong](https://cerebralvalley.ai/u/petroshong)
- **Demo video:** https://drive.google.com/drive/folders/14ovOTOBwFU4daupRUMiw_cwvMxOK4pJN
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=19

### 20. TOKYO DRIFT

I'm building an AI necklace that lets people interact with AI naturally through voice, making everyday tasks faster, more intuitive, and always accessible.

- **Team:** [Hotaka Funahashi](https://cerebralvalley.ai/u/Hotaka)
- **Demo video:** https://youtu.be/90PsuTeEpgE
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=20

### 21. FSD voice

When you are driving you either have a manual option or FSD. What you need is something in between to steer the car when needed - and voice is the best medium for this. You are stuck in a waymo and see a clear path - you should be able to give instructions instead of calling support. This is a POC demo to highlight such capabilities.

- **Team:** [Pradeep Banavara](https://cerebralvalley.ai/u/pradeep)
- **Demo video:** https://youtu.be/arvkGQASz9o
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=21

### 22. Voice Arena

Voice Arena is a realtime healthcare voice-tool firewall that demonstrates how voice agents with tool access can leak sensitive patient data when raw tool responses reach the model. It solves the gap left by prompt-only guardrails, which still expose secrets to the model and ask it not to repeat them, by enforcing a strict runtime data boundary so that unauthorized protected healthcare information (PHI) never enters the model’s response context.

- **Team:** [Yi Zu](https://cerebralvalley.ai/u/yizucodes)
- **Demo video:** https://www.loom.com/share/5575a8464132400b8f2894c72bda475b
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=22

### 23. CoverKey

Privacy first aggregator of property/auto insurance quotes.

- **Team:** [Igor Okulist](https://cerebralvalley.ai/u/okigan)
- **Demo video:** https://youtu.be/FOzl81dgtLI
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=30

### 24. FunPostcard.com

FunPostcard -- dictate your postcard and send it for $5

- **Team:** [Adam Koszek](https://cerebralvalley.ai/u/wkoszek)
- **Demo video:** https://youtu.be/o4AFOzw4geM
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=37

### 25. Clue

Clue Night is a realtime voice murder mystery where five gpt-realtime-2 agents — each a Clue suspect with private knowledge, motive, and something to hide, are interrogated by a human detective. One is the murderer. The agents reason in realtime, lie convincingly, react with shifting emotions (their voices modulate as composure cracks), and decide on their own when to interrupt each other through reasoning-driven tool calls, not server VAD. Explores how multi-agent voice systems behave under adversarial pressure and how realtime models handle deception as a sustained, multi-turn act.

- **Team:** [Aditya Suresh](https://cerebralvalley.ai/u/adi_suresh01)
- **Demo video:** https://youtu.be/oT6lDC9lnMg
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=43

### 26. Robert Huynh

Ăn Gì? is a cultural interface for unfamiliar food. It helps people cross language, dietary, and cultural gaps at the exact moment they need to decide what to eat via a realtime voice agent.

It solves a real problem for millions of immigrant, non native english speakers running mom and pop shops.

- **Team:** [Robert Huynh](https://cerebralvalley.ai/u/roberthuynh)
- **Demo video:** https://www.loom.com/share/d0be075ca395467ca8cad8c2554fd91e
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=23

### 27. emergency

help emergency responders communicate clearly and effectively without latency, maximizing potential of saving lives.

- **Team:** [Sanskar Thapa](https://cerebralvalley.ai/u/sanskar1016pro), [Pranav Sankar](https://cerebralvalley.ai/u/pranavsankar2)
- **Demo video:** https://youtube.com/shorts/v0g_zVsPRMU?feature=share
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=24

### 28. sayit

sayit is a voice-to-canvas workspace that lets users speak ideas and instantly turn them into visual artifacts: diagrams, SVG explanations, notes, and searchable local knowledge. It solves the problem of slow, manual whiteboarding by using realtime voice commands to create, update, connect, and save visual work in one live workspace.

- **Team:** [Yahya Alhinai](https://cerebralvalley.ai/u/yhinai)
- **Demo video:** https://youtu.be/NHeFJLDb6pQ
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=25

### 29. Dazzle

Processing live streamed earnings calls in real-time to extract financial data as well as semantically trigger hooks/tools whenever certain events occur in the call such as areas of concern or surprises.

- **Team:** [John Sabath](https://cerebralvalley.ai/u/johnsabath)
- **Demo video:** https://screen.studio/share/rJzitqum?private-access=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJ2ZXJzaW9uIjoiMi4wIiwic2hhcmVhYmxlTGlua0lkIjoiNzQ2NGEwZjItYTlkZi00NDAyLWI3ZGItYWRlZjg0ZWE1MmM1IiwiaWF0IjoxNzc5OTMxNzc2fQ.NBQE9_ab7dfbkZ1O8ZSjc5YwEfgvbkdknCbrwZR2H8A
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=26

### 30. Health Passport

Voice-first health copilot that runs on your body powered by the Meta Raybans with an in-lens display.

- **Team:** [Carl Vincent Kho](https://cerebralvalley.ai/u/Carl_NotANerd)
- **Demo video:** https://www.youtube.com/watch?v=nOmOFmkr5FY
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=28

### 31. Automat

Max is a fully-employed AI agent at Automat — he has a phone number, a Slack handle, a Gmail inbox, and access to the tools his teammates actually use. During today's testing, Pablo called Max from backstage and asked him to book dinner with whoever was available at the office. Max pulled cuisine preferences from Notion (Pedro likes paella, Pablo hates seafood), drove a real OpenTable reservation via headless browser, generated a personalized dining image from Slack avatars, and sent confirmations to Slack + email — all live, no scripts. 

Max later calls Pablo to let him know another person's flight was cancelled and were now available, to see if we wanted to update the reservation. we agreed and he sent through the updated reservation link. He sends all of us confirmation emails/slacks/and images, but it's hard to capture on video since there's actual computer use actions in between which were making the videos too large to upload in time. much better as a live demo with improvisation and crowd involvement

- **Team:** [Pedro Martinez Lopez](https://cerebralvalley.ai/u/p3droml), [Pablo Lleras](https://cerebralvalley.ai/u/Plleras), [Gautam Bose](https://cerebralvalley.ai/u/gbose)
- **Demo video:** https://youtu.be/Dd2lwC0F4e0?si=AP_BxB_uLuIixKCO
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=34

### 32. OpenSurgical.Ai

The Problem — the surgeon's hands are locked in
Robotic and minimally-invasive surgery solves many physical limitations of open procedures, but creates a paradox: the surgeon gains precision but loses access. For hours at a time, their hands are locked on the da Vinci controls inside a sterile field. Every critical piece of information — patient labs, CT scans, drug safety, complication protocols, anatomy — is one broken scrub or one distracted circulating nurse away.

OpenSurgical.Ai 
The surgeon's hands stay on the controls. OpenSurgical listens continuously, pulls the chart, navigates imaging, runs the safety timeout, flags drug contraindications, tracks blood loss, and writes the operative note — all from natural speech, in real time, without a single click.

- **Team:** [Bharat Bhavnasi](https://cerebralvalley.ai/u/bvsbharat)
- **Demo video:** https://youtu.be/1nf6fUr9cSg
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=35

### 33. Stella Foster

Stella Foster - an AI assistant available over the phone number - works even on the landline - and is capable of ordering Uber rides, purchasing things on Amazon Prime, generating images, texting and calling others, as well as scheduling reminder

- **Team:** [Illya Bakurov](https://cerebralvalley.ai/u/ibakurov)
- **Demo video:** https://youtu.be/Vg3E2rx8pzY
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=27

### 34. Scriber

Scriber is a two-agent voice architecture for
  real-time standup facilitation.

  Every existing meeting AI (Spinach, Otter, Fireflies,
  Fellow, MeetGeek) is passive-then-active: it listens,
  then post-meeting it acts. Scriber flips this — the
  agent acts during the conversation, with the Linear
  board visibly mutating live.

  Scriber (facilitator) joins standup as a voice agent.
  She prompts each teammate, listens for blockers and
  commitments, and executes Linear tools as words come
  out — creating tickets, marking states, attaching
  AI-generated diagrams via gpt-image-1.

  Mnemo (silent supervisor) runs in parallel on a
  deep-reasoning model. She holds long-term
  cross-standup memory and whispers tactical context
  when Scriber asks — citing dated prior commitments,
  repeated blockers, and threads the team is missing.

  The two-agent split decouples conversational latency
  (Scriber, low reasoning) from analytical depth (Mnemo,
   deep reasoning with full history) — a primitive
  nobody has shipped: a two-agent voice architecture
  where one agent runs the meeting and another remembers
   everything across meetings.

  The "visceral moment" of the demo: Scriber generates a
   diagram via gpt-image-1 during the conversation, the
  diagram attaches to a Linear ticket, and the audience
  sees the attachment appear on the ticket in real time.

  Built solo in ~3 hours. Repo:
  github.com/benikigai/scriber

- **Team:** [Benjamin Shyong](https://cerebralvalley.ai/u/BenjaminBear)
- **Demo video:** https://youtu.be/unZ0P_VZAoM
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=29

### 35. Sketchers

Forensics Drawer is a voice-guided forensic sketch prototype that helps capture a witness description, structure it into a suspect schema, and generate sketch iterations from the collected details. It uses OpenAI Realtime voice models for the interview flow, with Whisper transcription configured via whisper-1. It uses GPT Image models, currently selectable as gpt-image-1, gpt-image-1.5, or gpt-image-1-mini, to generate the suspect sketch outputs.
https://github.com/tanaypai/openai-voice-hack

- **Team:** [Tanay Pai](https://cerebralvalley.ai/u/tanaypai), [Kaustubh Gaikwad](https://cerebralvalley.ai/u/Kguy)
- **Demo video:** https://drive.google.com/file/d/1tYsh_ApmiIbMcr9MEMGrcAQu17bkVEFI/view?usp=sharing
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=31

### 36. KT languages

AI Language Learning tutor

- **Team:** [KT Yang](https://cerebralvalley.ai/u/ykaitao)
- **Demo video:** https://youtu.be/0KkAAXdvECo
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=32

### 37. wonderworld

Our core tech is a **Voice-to-World Engine**: a realtime system that turns spoken intent and live audio signals into validated operations on an interactive simulation.

WonderWorld is just the first use case. The underlying primitive can power education, scientific simulations, medical explainers, design tools, training environments, and any interface where users should manipulate a live model by speaking naturally.

- **Team:** [Rajashekar V](https://cerebralvalley.ai/u/raj), [Dien Hu](https://cerebralvalley.ai/u/de_an_hu)
- **Demo video:** https://youtu.be/hiH8I-7qqng
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=36

### 38. Aila

This demo stress-tests a real-time voice AI system in a
  high-pressure travel disruption. The user is at the airport,
  their flight is delayed, they may miss a connection, they
  have checked luggage, and they need help immediately.

  Instead of one model simply answering a question, this app
  shows multiple AI capabilities working together at the same
  time: real-time voice, transcription, memory, correction,
  autocomplete, intent prediction, safe tool prefetching,
  latency tracking, emotion signals, and evals

- **Team:** [Sahn Alam](https://cerebralvalley.ai/u/sahn)
- **Demo video:** https://youtu.be/tj0jtjt4EbY
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=38

### 39. Prism Voice

Prism Voice is a real-time AI voice filter that preserves your intent while turning raw speech into clearer, warmer, or more confident words delivered in a voice that sounds like your own.

- **Team:** [Ali Jeffrey Razfar](https://cerebralvalley.ai/u/arazfar)
- **Demo video:** https://www.youtube.com/watch?v=nb5lFmGLDQc
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=33

### 40. Codetalker - Voice OS for parallel AI coding agents

CodeTalker: Give Codex a voice! 

CodeTalker a macOS side panel that turns each AI coding session into a voice channel. CodeTalker uses Hooks to speak events so you can response verbally in realtime! Hold a button and address any agent by name or context ("Quincy, what's blocking you?" or "on the backend session, what's the latest?") while your eyes stay on whatever you're actually working on. Agents speak back in their own voices when they finish, fail, or need a decision. Voice isn't replacing the keyboard, it's a third channel that doesn't compete with eyes or hands.

- **Team:** [Ankit Tandon](https://cerebralvalley.ai/u/ankit_tandon), [Peter Allport](https://cerebralvalley.ai/u/peterallport), [Kai Brokering](https://cerebralvalley.ai/u/kaib), [Jonah Daian](https://cerebralvalley.ai/u/jonahdaian)
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=40

### 41. Miraii AI

Ai that calls your phone and talks to you 
Can we like a mentor. Strategist, friend, love
Checkout callmiraii.com to try demo
An extension to wearables would be Voice AI smart rings,  checkout miraii.ai for demo video

- **Team:** [Anu Bhat](https://cerebralvalley.ai/u/anubhat)
- **Demo video:** https://youtu.be/N2JF0BKOsj8?si=TF97ytyZXYDZtmR_
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=41

### 42. A Link to the Chat

GPT Realtime plays Super Nintendo games and provides Twitch TV style commentary as it solves puzzles

- **Team:** [Alex Reibman](https://cerebralvalley.ai/u/Alex), [Travis Cline](https://cerebralvalley.ai/u/tmc), [Sarath Shekkizhar](https://cerebralvalley.ai/u/shekkizh)
- **Demo video:** https://youtu.be/OtC1J443htw
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=44

### 43. Peripheral

A pair of smart glasses designed as a dedicated peripheral for personal AI agents.

The device weighs 28 grams and uses binocular waveguide optics with 540x280 micro-LED displays and 12+ hours of battery life. The software layer connects to the user's existing agents (Hermes, openclaw, Codex etc.) and renders agent state, output, and approval requests directly in the lens. Voice input is handled on-device and routed through GPT-realtime-whisper.

The system is built around the idea of agent-generated UI. Rather than shipping a fixed interface, Codex decides per-request how to display information: for low-latency tasks it uses pre-built templates, and for richer or one-off interfaces it calls GPT-image-2 to generate custom UI sized for the display. A skill-file layer tells Codex which models and MCP tools to use for a given task type (translation, summarisation, status checks, etc.).
Two core problems it addresses: (1) reducing friction between the user and their agents by removing the phone-unlock-and-open-an-app loop, and (2) surfacing remote agent approvals so coding agents don't stall when the user is away from their laptop. Live use cases shown include real-time translation, on-demand reference UI, and at-a-glance multi-agent status monitoring.

- **Team:** [Karim Yahia](https://cerebralvalley.ai/u/Kariminal)
- **Demo video:** https://www.loom.com/share/cca94191ca4d42f28aa9fd49cffd6abc
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=39

### 44. Athens

An embodied robot, the robot can walk, jump, and collect orbs in a secret world. Listens and replies with voice. Try it at https://snowman-taupe.vercel.app/ with full audio, Press t to talk and release mic in noisy environments.

https://github.com/ziadgit/snowman

- **Team:** [Ziad A](https://cerebralvalley.ai/u/ziad)
- **Demo video:** https://www.loom.com/share/64cf4359f94245828e2cbe78fd66bf1b
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=45

### 45. Ensemble

I did a Voice to 3D asset generation that also can create a video in any scenario lowering the creative barrier for folks to make movies, marketing material, or any embodied AI.

- **Team:** [Brandon In](https://cerebralvalley.ai/u/brandonin)
- **Demo video:** https://youtu.be/UsI_GZ6EGqc
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=42

### 46. Flex_care

Clinical EHR navigator - browser-based prototype for a realtime voice navigation layer inside healthcare workflows. It is designed for moments when a clinician, medical assistant, prior authorization coordinator, or care team member is stuck inside a dense EHR screen and does not know what to do next.

Instead of leaving the workflow to ask IT, message the Epic team, search a tip sheet, or file a support ticket, the user activates the voice assistant and speaks naturally. The agent then guides them step-by-step on the same screen, points to the relevant field using a floating coach-mark card, explains why something is blocking the workflow, and creates a safe escalation summary if the issue cannot be completed.

- **Team:** [Srushti Sunil Madhure](https://cerebralvalley.ai/u/Srushti247)
- **Demo video:** https://youtu.be/6RDHrZx-EeU
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=46

### 47. Raven

Raven is an AI reading coach for dyslexic kids.

- **Team:** [Caroline Dahllof](https://cerebralvalley.ai/u/cdahllof)
- **Demo video:** https://youtu.be/EaLcWxS-Rd8
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=47

### 48. Workview Pro

A voice agent for HVAC and plumbers that fills out all of their paperwork for them as they go about their work day.

- **Team:** [Armando McIntyre-Kirwin](https://cerebralvalley.ai/u/odnamra)
- **Demo video:** https://youtu.be/QSMEiHz9N1w
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=48

### 49. Agentic OS for a Phone

All built just today. The Next Phone is a voice-first mobile OS. You talk, it answers, takes action, and builds the right interface in real time: calendar, tasks, flights, email, news, and more. No app hunting. Just ask.

- **Placement:** 1st Place
- **Team:** [Isa Usmanov](https://cerebralvalley.ai/u/sintem)
- **Demo video:** https://youtu.be/x0C0etsyO0U
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=49

### 50. Wagner

Wagner is an interactive, multi-agent virtual meeting space powered by OpenAI's Realtime API that bridges the gap between technical execution and financial strategy. The platform brings two distinct AI agents into a live voice call—a DevOps Lead and a CFO—each equipped with unique, domain-specific contexts regarding a proposed infrastructure shift. As they actively debate the move in real time, the agents don't just speak; they use tool calling to dynamically surface visual artifacts, with the DevOps agent pulling up architectural diagrams and the CFO agent rendering live budget breakdowns. This creates an immersive sandbox for engineering and finance teams to stress-test decisions, visualize bottlenecks, and align on costs before committing resources.

- **Placement:** 2nd Place
- **Team:** [Yeferson Pena](https://cerebralvalley.ai/u/Yef), [Jhon Enciso](https://cerebralvalley.ai/u/Johncito), [Steve Suarez](https://cerebralvalley.ai/u/iamstevesuarez)
- **Demo video:** https://youtu.be/vwMd2znrUII
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=50

### 51. Pulley

Github PR explainer for product managers and non-technical stuff. Claire will walk you through all the updates to help with communication between Product leads and technical staff!

- **Team:** [Jonathan Belay](https://cerebralvalley.ai/u/jonbelay1)
- **Demo video:** https://youtu.be/e0MJQe6B5sE
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=51

### 52. Juno

AI health assistant for the 1B people living with chronic illness

- **Team:** [Marshall Gould](https://cerebralvalley.ai/u/MarshallGould)
- **Demo video:** https://youtu.be/HWiTI2-6hok
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=52

### 53. Live translate on Meta Glasses

Used the gpt-realtime-translate model to live translate whoever is talking to me on meta glasses-- much faster than built in, more languages, and works fine even when someone is videocalling me on a laptop. Also swapped out our LLM in Yelp's voice AI to use realtime-1.5 to reduce latency 200-500ms.

- **Team:** [Thavidu Ranatunga](https://cerebralvalley.ai/u/serene)
- **Demo video:** https://drive.google.com/file/d/1XUF39mQpx5HJINHHPDIlq8ZrefLjj_JE/view?usp=sharing
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=53

### 54. Curo

Curo gives every kid their own one‑on‑one physics tutor: a cute AI buddy that teaches first principles out loud and (socratically), taking real aim at Bloom's 2‑sigma problem. It talks with the kid in real time, draws a simple GPT‑image‑2 picture of each idea (gravity, magnets, the solar system) on a shared whiteboard, and the kid writes their answer right on that board for Curo to see and react to. Built end‑to‑end with Codex, powered by OpenAI Realtime Voice, KateX (for the whiteboard) and GPT‑image‑2.

- **Placement:** Finalist
- **Team:** [Ansh Chopra](https://cerebralvalley.ai/u/ansh13)
- **Demo video:** https://youtu.be/V0d2ivQzpm4
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=54

### 55. FileVine Realtime Voice

Built a real-time voice AI interface for Filevine using GPT Realtime 2 and a custom MCP server. The project lets users talk directly to their Filevine data through natural conversation, bringing agentic voice workflows to a platform that previously had no MCP integration.

- **Team:** [George Pickett](https://cerebralvalley.ai/u/Itsgeorgep)
- **Demo video:** https://www.loom.com/share/aed419eafc934647ae276d2a10160c22
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=55

### 56. The Shrine

A voice-first AI therapist where you share what brought you, get matched to a real Japanese temple suited to your concern, walk through a poetic ritual, and leave with a downloadable amulet and a personal blessing.

- **Team:** [Iris Wang](https://cerebralvalley.ai/u/kmwzs66)
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=56

### 57. VoxTerminal

VoxTerminal is a voice-driven command center that empowers developers to direct a team of parallel, autonomous coding agents.
Unlike traditional black-box sub-agents, it allows real-time interruption, individual addressability, and instant course-correction during execution.
Developers can dynamically spawn new agents or pivot existing tasks entirely hands-free, preserving their coding flow and context.
It fundamentally shifts the developer-AI relationship from linear, turn-based chatting to dynamic, parallel directing.

- **Team:** [Tatuki Yabe](https://cerebralvalley.ai/u/yabecchi)
- **Demo video:** https://drive.google.com/file/d/1HO9CiO_-PmCoQsBCxUBvGLNrOv8JiXvB/view?usp=sharing
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=57

### 58. Realtime Workout Coach

Realtime OpenAI voice workout coach that guides exercise, logs structured state, runs background timers, replans when time changes, and adapts to discomfort.

- **Team:** [Rupert Dodkins](https://cerebralvalley.ai/u/Rupert)
- **Demo video:** https://qaby.ai/short/1X6JGC
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=58

### 59. Gallatin AI

We’ve built from scratch a radio agent that is always in the loop.

It listens to all available channels, transcribes, categorizes and acts, fully autonomously.

Integrated for both green gear and more advanced MANET radios. 
We built this across the Maven Smart System and Navigator, a leading tactical resupply software. We are Gallatin, a defense logistics startup.

- **Team:** [Yizel Vizcarra](https://cerebralvalley.ai/u/ychats)
- **Demo video:** https://www.loom.com/share/f7ec815ef5b04802b9cd3468d408dca4
- **Project:** https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery?project=59

---

Markdown version of https://cerebralvalley.ai/e/openai-voice-hack-night/hackathon/gallery. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt