Tavus

On March 26, a 74-year-old cancer survivor named Jerome~W Dewald stood before a New York appellate court… sort of. Instead of appearing in person or even on Zoom himself, he used an AI-generated avatar named Jim—a clean-cut, confident digital spokesperson he created through Tavus. Why? Because extended speaking is physically difficult for him. He had prior approval to present remotely. But no one expected a synthetic human to show up. The video barely lasted a few seconds before the judges shut it down. Dewald was forced to speak live—the very thing he was trying to avoid. In an exclusive interview with us at AIM Research, Tavus Co-founder Quinn Favret told us: “Jerome used an AI-generated video because he has speech challenges and wanted to present a professional and audible statement in court.” This wasn't about trying to deceive anyone. It was about accessibility. It was about a man with real limitations trying to find a modern workaround. As AI tools become more human, maybe the systems they enter need to become more humane. Check out the link in the comments to read more 👇 Hassaan Raza 

We’re building the next form factor of AI. Come shape it with us 👇

𝐓𝐚𝐯𝐮𝐬 𝐰𝐞𝐧𝐭 𝐭𝐨 𝐜𝐨𝐮𝐫𝐭... 𝐚𝐧𝐝 𝐦𝐚𝐝𝐞 𝐥𝐞𝐠𝐚𝐥 𝐡𝐢𝐬𝐭𝐨𝐫𝐲 (𝐬𝐨𝐫𝐭 𝐨𝐟) 🤯 Last week, something extraordinary happened in a New York courtroom. A 74-year-old man named Jerome Dewald walked in and pressed play on a video. On screen: a calm, articulate “lawyer” presenting Jerome’s oral argument. Only it wasn’t Jerome. And it wasn’t a real person. It was an 𝐀𝐈 𝐯𝐢𝐝𝐞𝐨 𝐚𝐠𝐞𝐧𝐭, created using Tavus. Jerome, who struggles with verbal fluency, used our tech to express himself clearly and confidently. He had permission to use the video. ...but the judge was furious, “I don’t appreciate being misled,” she said. The hearing ended abruptly. But here’s what the media missed 👉 Jerome wasn’t trying to hack the legal system. He was trying to be heard. The story went viral. Hundreds of thousands of people chimed in, with comments like: 🗣 “An AI lawyer with access to every precedent and case law might be really effective.” 🤖 “I wanna see an AI lawyer vs AI judge.” 🗣 “I’m more surprised a 74-year-old man knew how to generate AI.” The engagement was massive. It struck a nerve. Not because it was absurd, but because it felt inevitable. At Tavus, this wasn’t just a tech moment. It was a 𝐠𝐥𝐢𝐦𝐩𝐬𝐞 𝐨𝐟 𝐭𝐡𝐞 𝐟𝐮𝐭𝐮𝐫𝐞 we’re building. We’re not here to replace humans. We’re here to amplify them. Our AI replicates face, voice, expression, nuance in real-time–the essence of human connection. That matters deeply in law, education, healthcare, and beyond. And Jerome’s legal use case? Just the beginning. Imagine this: 🔁 Mock trials with real-time AI judges 🧑⚖️ Interactive legal reps for those who struggle to advocate for themselves 🤝 Conversational legal guidance before hiring a lawyer Not science fiction. Just good tech, built with transparency and intent. The courtroom wasn’t ready. But the future knocked. Jerome answered. Next time, let’s not slam the door. Let’s open it—and step forward, together. #AI #LegalTech #JusticeForAll #Accessibility #FutureOfWork #Tavus #HumanConnection #SpeechAccessibility

Congrats Abhishek Mohanty & Careflick! Incredible use case of using video agents to provide companionship for the silver years 👵 ❤️

Building AI Agents in Health Care? Learn how to make your agent look and feel human using Tavus conversational video APIs with Karthik Ragunath Ananda Kumar this Wednesday in Palo Alto.

Exciting to see so many companies on this list that are powered by Tavus 😍 🤖

Big Tavus shoutout, we’re featured on the Meta blog! 🚀 We’ve been heads down building the most natural, real-time digital conversations out there. Our Conversational Video Interface (CVI) blends video with cutting-edge perception and reasoning, powered by Llama 3.3. With Llama, our digital replicas engage in dynamic, multimodal conversations that adapt to context and feel truly lifelike. Why Llama? 🔹 Richer, more natural dialogue 🔹 Lightning-fast, real-time performance 🔹 Open-source flexibility + privacy control 🔹 Deep context understanding with RAG customization And a massive thanks to Cerebras Systems and Fireworks AI for helping us achieve up to 550% latency improvements over traditional pipelines. 🧠 Check out the full write-up on how we’re using the latest Llama models to push conversational AI forward: https://lnkd.in/gknW4zAE #AI #OpenSourceAI #Llama3 #Tavus #ConversationalAI #CVI #VideoAI #MultimodalAI

I noticed there’s something missing from most AI conversations today... They understand words. They sound kinda human. They even call functions and juggle tools. But I still don’t think they feel quite right. If 2024 was about making conversational AI work. For us, 2025 is about making AI conversations feel human. Most conversational AIs wait ~1.3 seconds after you stop talking before replying. That silence is awkward. It causes interruptions, confusion, and leads to worse outcomes. At Tavus our research team built Sparrow-0 to fix this problem. Instead of waiting for silence, Sparrow, our new turn-taking model, predicts the right moment to speak based on meaning, tone, and context, just like humans do. We went deep on how people actually converse and found that humans don’t actually wait for a pause, they’re constantly processing both what’s being said and how it’s being said. That’s how we know when to jump in. Sparrow simulates this by analyzing semantics and context cues in real time to respond more naturally. This timing isn’t just a UX detail. It’s a trust signal. When AI speaks like a human, people relax. They stay engaged. Sometimes, they even forget it’s AI. When we dug into the data, even we were surprised to find that with Sparrow there were: ➡️ Nearly 2x faster replies ➡️ 50% fewer interruptions ➡️ 80% higher retention This stuff really blew our minds while building it! If you're interested in learning more, we just published a deep dive on how Sparrow nails the timing to make AI feel less like a robot and more like a real conversation.

Introducing: Pigeon-0, our SoTA model for taste and smell 🐦 At Tavus, we’ve always believed AI should see, hear, and now—smell and taste. That’s right, our latest model gives AI replicas the power of olfactory and gustatory senses. What is Pigeon-0? A state-of-the-art olfactory-gustatory transformer model trained on a blend of open-source Google Nose data and unlabeled subway footage. Pigeon-0 allows your AI replica to detect, identify, and even emotionally react to smells and tastes in real time. Just a few conversational video examples made possible by Pigeon-0: 🍷 AI sommelier: Instant judgment of your wine choices ❤️ Matchmaking assistant: Pheromone-based dating 🍗 Food safety monitoring: Never trust questionable leftovers again Built on the shoulders of giants: A big thanks to the Google Nose team (2013–2013) for open-sourcing the revolutionary scent model that we now leverage. 

Congrats on the launch Alfie Whattam & Alfa AI! Excited to build together 🦾