Geoff Hulten

Podonos just raised $2.4 million in pre-seed funding, and the signal is loud and clear: the voice AI industry finally has someone measuring the sound, not just making the noise. Based in Los Gatos, Podonos is building the infrastructure layer that decides whether voice AI feels real, human, and ready for primetime. The industry is racing toward a projected $47.5 billion by 2034, but growth without standards is chaos. Podonos is making sure the hype actually holds up when you press play. At the center is ✦Soohyun Bae, PhD in Computer Engineering from Georgia Institute of Technology and Y Combinator W22 alumnus. His track record stretches from engineering roles at Google Maps to leading AR mapping at Niantic, plus co-founding Bobidi and TickTock AI. He has seen what happens when technology scales without proper #guardrails. With Podonos, he's betting the future of AI isn't just about how smart the models get, but how believable and trustworthy they sound. Podonos evaluates models across naturalness, similarity, emotion, recognition accuracy, pronunciation, tone, and resilience. Add in #personaconsistency, and suddenly you're not just testing machines, you're testing their ability to pass as human. The twist? They deliver results in under twelve hours, when the legacy approach drags on for two months. In AI time, that's the difference between relevance and irrelevance. Investors caught the beat. Serac Ventures, led by Kevin Moore, took the lead, backed by NAVER D2SF, the venture arm of South Korea's NAVER, and KAIST Venture Investment Holdings. Together, they've now backed Podonos twice, with this $2.4 million round stacking on top of a $750,000 raise in March 2025. Total funding sits at $3.15 million, and every dollar is pointed toward scaling #engineering, refining #analysistools, and launching into the U.S., Europe, and Southeast Asia. This isn't theory. Podonos already serves Resemble AI, Play AI, and Sanas AI, hitting six-digit ARR by November 2024 with just four employees. Their secret weapon is reach: 150,000 evaluators spread across nine languages and thirteen locales, forming a human-in-the-loop platform that blends speed, rigor, and global coverage. That reach lets Podonos deliver evaluations at a scale and pace that no competitor has matched. The strategy is ambitious: expand into high-demand sectors like #healthcare, #finance, #gaming, and #advertising, then move beyond voice into #multimodal model evaluation, spanning video AI and large language models. The mission isn't just to check if AI works, it's to decide if AI works for people. #Startups #StartupFunding #EarlyStage #VentureCapital #PreSeed #AI #AIInfrastructure #Voice #VoiceAI #VoiceTech #Infrastructure #Technology #Innovation #TechEcosystem #StartupEcosystem #Hiring #TechHiring If software engineering peace of mind is what you crave, Vention is your zen.

6

1 Comment

Open AI’s Codex 5.3 and Claude Opus 4.6 (released on Feb 5th) aren’t just incremental upgrades, they signal a structural shift. What’s emerging is not better content generation. It’s increasingly dependable multi-step reasoning inside real workflows. And that is a big jump. When models can plan, execute, and maintain context with growing stability, AI stops being a tool you use. It becomes a system you design around! #AI #GenerativeAI #LLMs

38

3 Comments

We released a new VibeVoice model, VibeVoice ASR, a unified speech-to-text model designed to handle 60-minute long-form audio in a single pass, generating structured transcriptions containing Who (Speaker), When (Timestamps), and What (Content), with support for Customized Hotwords. It redefines ASR in the era of LLMs, and the new ASR will be a foundation to unlock the potential of new AI-Native hardware, memory systems for AI, and many other scenarios. Please have a try at https://lnkd.in/dNCCzQiN Key features: -         60-minute Single-Pass Processing: Unlike conventional ASR models that slice audio into short chunks (often losing global context), VibeVoice ASR accepts up to 60 minutes of continuous audio input within 64K token length. This ensures consistent speaker tracking and semantic coherence across the entire hour. -         Customized Hotwords: Users can provide customized hotwords (e.g., specific names, technical terms, or background info) to guide the recognition process, significantly improving accuracy on domain-specific content. -         Rich Transcription (Who, When, What): The model jointly performs ASR, diarization, and timestamping, producing a structured output that indicates who said what and when. We are continuously pushing the frontier of voice AI. More release on the way. Stay tuned! https://lnkd.in/dATPu9nw

380

8 Comments

Conditioning with 3d bounding boxes is an intuitive step for editable scene generation.

26

Conversational Agentic Systems: Cursor's agent mode, Zed's agentic editing, and Claude's workflows are converging on a similar pattern — you can steer AI through conversation while it handles complex multi-step work. I wrote about my observations on this emerging trend the industry is seeing. https://lnkd.in/eDmXUveG

3

Big news for Meta Neural Band owners! 🎉 Starting this week, we're rolling out a groundbreaking new way to message: writing with your finger on any surface (https://lnkd.in/g6t_9sPC). Imagine messaging as simply as you wrote in elementary school—no keyboards, no accessories, just you and the neural band. This isn't just a new feature; it's a fundamental step in the evolution of human-computer interaction. The neural band natively understands written text using human biology (EMG at the wrist), eliminating the need for users to learn a new interface (physical/virtual keyboards, touch screen UIs or mice). This is a new step in the paradigm shift from humans adapting to computers to computers communicating like humans. While much of the recent AI focus has been on language models and reasoning, the same underlying techniques have been taking perceptive AI to quite a mature level over the last 10 years. First it revolutionized computer vision (https://lnkd.in/gejhZ9Vu) and speech recognition (ASR, TTS). Beyond human senses, it enabled the processing of complex sensor suites that made applications like self driving cars possible (https://lnkd.in/ghxE2hvR). When we launched the Neural Band last fall (https://lnkd.in/gyuzRVeD), we proved for the first time that a consumer-grade device could accurately capture and process neural activations. Initial gestures were innovative, but handwriting is different. While finger differentiated taps and swipes could have arguably been implemented using other modalities, the richness and accuracy of the signal required for handwriting requires the high-information-entropy bandwidth unique to an EMG-based device. I can't wait for many of you to experience the "waouh, this is new!" feeling for yourself. While this is the culmination of 10 years of research (https://lnkd.in/gxm_gc3z), we're just getting started! Stay tuned for the even more exciting, next-level capabilities we are working on. #MetaNeuralBand #HumanComputerInteraction #PerceptiveAI #Innovation #EMG

711

22 Comments

Couldn't agree more with Anand - while I understand the skepticism and concern people have with this approach, it has utility and value for many things. I still say this is just "software development" in a new age, and that the term "vibe coding" diminishes the effort involved in making it work. There's a process and discipline to doing this - but it can work quite well. Whatever you call it - if you aren't learning how to do this, you are going to get left behind... quickly.

2

11 Comments

AI-tools are everywhere, but they require thoughtfulness, creativity, and grit to use well. Josh continues to blaze the trail on this at Encamp. If you're interested in getting the most of AI development in an existing and fast-moving code base, I highly recommend reading this post.

9

The evolution of segmentation models has been pretty wild. Especially SAM. What felt experimental not long ago is now genuinely usable. SAM-3 is getting close to human-level performance in a lot of real-world cases. That changes the math: -Less manual labeling -Faster iteration -Smaller teams doing work that used to require entire pipelines The interesting question isn’t “can the model segment this?” anymore. It’s “what systems can we now build because this is reliable?”

14

This is what a good GTC week looks like. Spectro Cloud PaletteAI hit GA, and with it came validated integrations with WEKA, Aviz Networks, Netris, F5, and NVIDIA's own reference architectures like AI Grid and DSX Air. What ties it all together is the operations layer. Every one of these partnerships solves a real day-2 problem: storage performance, network automation, tenant isolation, security without throughput loss. These aren't just logos on a slide. They're tested, integrated, and ready to run through cluster profiles. Swipe through the carousel for the quick version, or hit the links for the deep dives.

1