Offir Gutelzon

We're back with another Startup Radar post highlighting up-and-coming early stage Seattle-area tech companies. Vercept, led by Kiana Ehsani, is building a new computer interface with help from AI and just released more details about its product. CHASSY, led by Chuka Okoye, just raised $1.2 million to help companies operate software for robotic fleets. Haply, Inc., led by Prasad Anguluri, helps Homeowners Associations manage resident requests, policies, and maintenance. Context Nature, PBC, led by Ephi Banaynal dela Cruz, helps finance and tax professionals identify clean energy tax credits. Axtrolite, led by Atul Alatkar, is a stealthy startup in the real estate property management sector. Read more about the startups and founders: https://lnkd.in/gEf4agqj

128

9 Comments

With RL techniques like GRPO, it’s become easy to fine-tune small models to outperform foundation models on vertical tasks. We’re open sourcing Osmosis-Apply-1.7B: a small model that merges code (similar to Cursor’s instant apply) better than foundation models. Links in comments! Using foundation models for high specificity, low complexity tasks like code merging can be overkill - instead, it’s better to use a specialized model instead for higher performance and latency. We fine-tuned the model on 100K real-world code commits where the model needs to merge a code edit snippet into an original code snippet. On the test dataset (subset of CommitPackFT), foundation model performance ranged between 0.77-0.93 reward score. In comparison, Osmosis-Apply-1.7B achieved a 0.98 reward score - while also being 3X-5X more affordable than the next cheaper model AND ~10X faster! Why this matters: the best AI agents use multiple models to ensure task specialization and reliability at scale - on top of cost and latency benefits as well. Links to the model and full write-up (with reward function, hyperparameters, etc.) are in the comments - reach out if you’re interested in using reinforcement learning!

239

18 Comments

We at Blitzscaling Ventures are proud to invest in Atrios, an a16z portfolio company where we saw CEO Taylor Offer demo the product at the a16z Speedrun Demo Day. Atrios is an AI-powered network intelligence platform that maps who knows who and makes those relationships actionable for go-to-market teams. How it works: -Tech vendors can post offers and specify the types of roles they want to connect with. -Connectors (often founders, VCs, business development professionals between roles, or highly connected individuals) can peruse the offerings and choose the best fits. -The connectors reach out to their prospects for a mutually agreed-upon introduction. -The tech vendor gets a qualified, warm intro to a lead. The connector gets compensated. The business decision-maker accelerates their decision-making process. -All three parties benefit: the vendor, the connector, and the business decision-maker, which is why the company is cleverly called “A-Trios,” Atrios. And yes, Taylor Offer that's his actual given name, perfect for using AI for modern business development. Link to our post in comments

110

12 Comments

We at Sound Ventures are incredibly excited to share today we led WitnessAI's $58M raise, with participation from Fin Capital, Qualcomm Ventures, Samsung Ventures, and Forgepoint, alongside existing investors Google Ventures and Ballistic Ventures… And it is just the beginning 🙌 The primary barrier to enterprise AI adoption isn’t tech debt; it’s tech doubt. WitnessAI has proven it can instill confidence in AI adoption, and has done so for multiple global leaders. We are thrilled to support the team as they use this capital to expand into new markets and solidify their position as the leading standard for safe, secure enterprise AI. In our diligence, we spoke with dozens of Fortune 500 security leaders and heard a clear theme: teams want to move fast with AI, but need protections that scale. WitnessAI provides that through a full, end-to-end security platform spanning AI applications, models, and agents, seamlessly integrating into a company’s existing infrastructure. As AI moves from experimentation to execution, WitnessAI enables organizations to deploy AI broadly and responsibly. Led by Rick Caccia, a world-class operator with over two decades of experience building and scaling security businesses, WitnessAI is on a path to define the category for enterprise AI security.

476

38 Comments

Remilk is here 🥛 It took a few years (real innovation always does), but ‪Aviv Wolff‬‏, Ori Cohavi, PhD, and team have delivered: real dairy made without cows, produced at industrial scale and finally available for the end consumer. What began as a crazy, almost outlandish idea is now being poured in coffee shops and served across food service across Israel, with retail around the corner. This is a watershed moment for the category and a major milestone for the Israeli deep-tech ecosystem. Onwards! 🚀

260

22 Comments

There is so much magic for scientists locked away in LLMs. But unlocking it requires a rare blend of scientific and engineering expertise. That’s why I’m thrilled that Nicholas Larus-Stone and Sphinx Bio are joining Benchling. I’ve gotten to know Nicholas over the last few months. We share a vision for what AI agents can do for scientists – and an obsession with turning that potential into real, everyday value. Together, we’re building agents that automate toil and answer complex questions. See what we’re building: https://lnkd.in/gtDHZpnt

372

8 Comments

The best founders don’t just build products; they reframe the problem. Ryan Alshak has done exactly that with Laurel - turning time from a background task into a source of insight, alignment, and leverage. IVP is leading Laurel's Series C to help build the system of record for how work actually happens. Read more from Allie Garfinkle: https://lnkd.in/gweNiwsc

63

This woman built tech that can spot bipolar episodes, detect postpartum depression, and keep her diabetic son alive. Vivienne Ming is a theoretical neuroscientist, a serial entrepreneur, and a self-proclaimed “mad scientist.” Her obsession? Maximizing human potential. Vivienne was born Evan Campbell Smith. By her early twenties, she had dropped out of college, fallen into depression, and was contemplating suicide. Then she made a decision: if she was going to keep living, she was going to do something substantial. She flipped a coin between economics and cognitive science. It landed on cognitive science. She went back to UC San Diego, earned a degree in cognitive neuroscience, then completed a PhD in psychology and theoretical neuroscience at Carnegie Mellon. While in grad school, she met Norma, a fellow PhD student. They fell in love. And one night, Vivienne told her the truth. “My secret is that I wish I were a woman.” Norma didn’t run. She listened. Vivienne underwent hormone therapy after conceiving their first child, Baxter, and completed gender confirmation surgery in 2008. They later had a second child, Thalia. Over the next two decades, Vivienne launched six startups, and served as chief scientist at two others. She invented AI systems that could: - Predict manic episodes in bipolar patients - Reunite refugee orphans with surviving relatives - Help autistic children learn to read facial expressions - Manage her son’s diabetes in real-time - Identify and mitigate bias in hiring - Improve cognitive recovery in TBI and dementia patients - Assess student creativity and emotional intelligence In 2011, she co-founded Socos Labs, her fifth company, a “mad science incubator” focused on the intersection of neuroscience, AI, education, and human development. Then she built The Human Trust, a nonprofit data trust that works on complex social challenges: disability, economic exclusion, algorithmic justice… She also co-founded Dionysus Health, where she helped create the world’s first biological test for postpartum depression, using AI and epigenetics to identify risk before symptoms appear. Throughout it all, her central thesis never changed. AI is not the end of the world. It’s not salvation either. It’s a tool, it’s only as useful as the humans who wield it. The real danger, she says, is that we’ll use AI to do just enough. That we’ll automate away our creativity. Flatten our curiosity. Become passive operators of tools we don’t fully understand. When writing her latest book, Robot-Proof, she fed every chapter into Gemini. But not to edit. “I tell it: You’re my nemesis. Tear this apart. Tell me where I’m wrong. Then I run it again: Now you’re a bored reader. Show me where you got lost.” Her work has earned her a spot on the BBC 100 Women list, Inc. Magazine’s 10 Women to Watch in Tech, and more than a dozen research and innovation awards. 💡 Follow Justine Juillard for daily #femalefounder spotlights

360

24 Comments