Cartesia x Anthropic Voice Agents Hackathon hosted by Notion

Create a speech coach based on phoneme-level dysfluency detection.

Memento: Voice-Enabled Memory Care Assistant Memento helps individuals with memory loss maintain daily routines through empathetic voice interaction and visual monitoring. Our initial focus: medication management. What It Does Memento listens and watches. It processes conversations and camera feeds to provide timely reminders, verify correct medication selection, and detect emotional states for more compassionate interactions. All medication schedules and adherence logs sync with Notion. Technical Approach Built on Cartesia's voice agent platform, we extended their LlmAgent with: Custom tools for camera monitoring and Notion integration. An experimental wrapper architecture that processes images as continuous streams rather than tool calls. We explored two paradigms for multimodal agent design: discrete tool invocation versus continuous stream processing.

BrowserPowered-VoiceAssistant

An AI-powered sports commentary that adapts to every viewer, explaining rules for beginners, diving deep for experts, and celebrating your favorite players by name and interacting with YOU while watching the game,

Backsell Podcast ads today are broken. They are loud, awkward, and obviously shoved in at random moments , so listeners skip them every time. Advertisers waste money, and creators lose trust with their audience. Backsell fixes this. Podcast creators upload their episodes, browse a marketplace of available ads, and pick the ones that fit their content. Backsell's AI then does the rest: it listens to the entire episode, finds the perfect natural moment to mention the product, writes an ad script that fits the conversation, and delivers it in the host's own voice ,so it sounds like the host genuinely brought it up, not like a commercial was spliced in. For creators: Upload your podcast, choose an ad, get paid , no extra recording, no editing, no awkward ad reads.

Generate podcast-style ad reads from any YouTube video; Addy ingests a transcript, clones the voice, lets claude choose natural ad breaks, writes the ad copy, and voices it with Cartesia TTS.

Grape is a proactive product manager's assistant that helps you run background research, organize your work, and more so you don't have to.

Coding agents are a new form of entertainment, both for creators (right now), and for viewers (not as well served, mostly just look at X posts). People are already getting entertained watching agents on a social network (Moltbook virality). Let's give voice & personality to coding agents as they work into a new type of sporting event, with devs able to select coding agent "fighters" with different personalities / styles that compete on the same task and fight it out in a Street Fighter style battle, with the ultimate winner attaining eternal glory.

Table Read is an AI-powered script rehearsal app where actors practice scenes with an AI scene partner and receive real-time directing feedback. A human actor joins a LiveKit voice room and rehearses a scripted scene beat-by-beat with an AI character agent (powered by Cartesia Sonic-3 TTS and Anthropic Claude). A separate AI Director agent listens to each delivery in real time, evaluates whether the actor is truly acting or merely reading, and provides scoring, engagement analysis, and actionable notes. Scenes are managed via a Notion database and a Next.js frontend, and the final rehearsed script can be locked as a versioned artifact.

VoicePal lets users create a friendly persona that can call local businesses on their behalf (pricing, availability, bookings, etc.). Unlike typical voice bots, VoicePal is human-in-the-loop: at commitment points (booking, sharing contact info, confirming price/time), the agent pauses and asks the user for explicit approval inside the app. It solves the problem of making repetitive phone calls while keeping the user in control and producing structured outcomes (not just transcripts): what was agreed, next steps, and key details. VoicePal is designed as a PWA so it can be installed as an app on iPhone, Android, Mac, and can be used the website!

This project solves the problem of needing an always-available pair-programming partner by giving you a real-time voice assistant for coding challenges. It supports both autonomous execution (bot_drives) and coaching mode (human_drives), with struggle detection that triggers targeted hints when you stall, repeat failures, or ask for help. It also solves progress tracking by combining sandboxed test runs, level-based challenge progression, and a Notion session log of what happened and why.

Keto is a multi-modal voice AI agent featuring three interactive arenas that showcase different AI capabilities. The Banter Arena provides natural conversations with web search integration for real-time information. The Puzzle Arena offers 10 brain-teasing challenges with progressive hints and conversational solving. The Paint Arena enables AI-powered image generation through voice commands. Keto solves the problem of rigid, single-purpose voice agents by creating an expressive, situationally-aware AI companion that adapts to different contexts - whether you need current information, mental stimulation, or creative assistance. With interruption handling, personality customization, and real-time web search, Keto demonstrates how voice agents can be both functionally powerful and naturally conversational.

A vision of a screen-free voice-first family. Kids can get their questions answered by talking to the answering machine; a context aware AI that lives inside of a retro phone. Parents can check in on their kids interests. And in the background an agent uses Browser Use to place holds on books the kids are interested in.

Toast is a user testing stimulator voice agent that roasts your product ideas. Instead of slow user testing, Toast simulates distinct user personas, such as a confused Boomer Dad or a skeptical VC, pulled dynamically from a Notion database. Using Cartesia’s expressive voice AI and Anthropic’s reasoning, these personas react in real-time, delivering brutally honest, entertaining feedback on your product ideas before you write a single line of code. It turns the chore of user research into interactive, constructive conversation.

Obama cares is an AI-powered vishing simulation platform that uses Cartesia voice agents and Anthropic's Claude to conduct realistic, personalized social engineering phone calls for authorized corporate security awareness training. After each call, the system automatically analyzes the transcript, scores the trainee's vulnerability, and saves a full assessment report to a Notion database.

Amiga Voice Agent is a copilot for patient-doctor calls to help generate questions that the patient could not think of. Amiga has been 1st implemented to help fertility patients to understand their treatments and prompt question during their doctors appointments.

Prism is an AI-powered research IDE that conducts autonomous customer interviews via voice and surfaces structured product intelligence -- no PM needs to be on the call. A customer clicks a link, talks to an AI interviewer, and the PM opens Prism to find organized issues, personas, quotes, competitive intel, and analytics waiting for them.

voice AI for mortgage application search

StandupAI - Voice-Powered Daily Standup Assistant

Sales agent to collect leads and payments for my Shopify design, dev and consulting services.

Acco is a voice-first web safety agent that acts as a guardian for blind and visually impaired users browsing the internet. Users speak a goal like "Pay my electricity bill" and Acco navigates a real cloud browser, detects scams and risky pages in real-time, and guides users with expressive voice — adjusting tone, speed, and urgency based on what it sees. It features a dual-stage risk analysis pipeline (fast Haiku check + deep Sonnet analysis), domain verification to catch brand impersonation, payment confirmation gates with amount readback, and natural conversational voice responses powered by Cartesia Sonic. The agent runs on Browserbase with Stagehand for live browser automation, Claude for reasoning, and supports both push-to-talk and Cartesia Line for real-time streaming voice interaction.

Help me and friend plan a festival trip

Live link https://mediaagents.up.railway.app/ Follow the Money is a series featuring two (agent) hosts who investigate who owns corporate media. Discussing their findings on the money trail, ownership data, and biases of the top 6 publications, using real-time data to highlight current events and historical data for a full investigation. In this edition "Who funds corporate media?" with your favorite hosts (two agents) Andrew (investigative reporter) and FJ (industry veteran) investigate, discuss and debate their findings. What makes this special? - Characters have real personalities, are opinionated and converse naturally between each other - Conversations reference real-time events from this week via live web search (lots happened in the corporate media space this week if you follow the news like huge Washington Post layoffs) - Ownership data cross-referenced from curated dataset, Notion database and web - Bias and factuality ratings sourced from Ground News

TIC (Trend-Informed Creator) is a voice-first AI assistant for social media creators who want to plan and publish content without constantly chasing trends or juggling tools. Creators talk through ideas in natural conversation, and TIC helps research current trends, generate scripts, and organize posts. The agent uses real-time search to ground suggestions in what is trending right now and adapts recommendations for different platforms like TikTok, YouTube, and Instagram. It stores scripts, research, and schedules directly in Notion, maintaining a structured content calendar and long-term memory of creator preferences. By combining voice interaction, live trend awareness, and automated planning, TIC reduces friction in content creation and helps creators stay consistent and relevant with minimal effort.

Gathering data about how people define their community is a critical step in the redistricting process. In 2026, realistic voice agents with tool calling have made it possible to collect community data from the public with only a phone call rather than the clunky process of going to a website and drawing a complicated map. This will let everyone have their voice heard in the redistricting process.

Video Whisper transforms passive video watching into interactive conversations. Paste any YouTube URL, and we automatically clone the speaker's voice. Pause anytime to ask questions - the character responds in their own voice with context-aware intelligence. Educational videos preview future topics; anime characters genuinely don't know their fate. Save your Q&A moments as video clips directly to Notion!

Delegate is a voice-first agent that acts on your behalf in real-world tasks. After a short conversational intake, it searches for relevant providers, validates them using live web data, and reaches out with consistent questions (simulated for demo). It then returns a clear summary highlighting tradeoffs and uncertainty so the user can decide confidently. Delegate emphasizes delegation with control — nothing happens without confirmation, and the agent never makes decisions for the user.