?? Last Chance to Sign Up! ?? Join us for the second session of our webinar, this time with Sebastian Lozano from Redfin, where he'll dive into the development and evaluation of "Ask Redfin," a virtual assistant rolled out across 14 U.S. markets. In this session, we’ll cover key strategies to effectively evaluate your AI system, how to define success, and explore a range of evaluation metrics. When: Tomorrow ? Time: 1:00 PM ET ?? Sign up here: https://lnkd.in/dTPBcscz
关于我们
Vellum is a developer platform for building production-worthy applications on LLMs like OpenAI’s GPT-4 or Anthropic’s Claude. Use Vellum to save hours on prompt engineering, iterate on prompts in production confidently, and continuously fine-tune for even better results.
- 网站
-
https://www.vellum.ai/
Vellum的外部链接
- 所属行业
- 软件开发
- 规模
- 11-50 人
- 总部
- New York
- 类型
- 私人持股
- 创立
- 2023
地点
-
主要
获取路线
Madison Avenue
US，New York
Vellum员工
动态
-
Vellum转发了
The Vellum team is at THREE different events today, Miami, New York and hosting our virtual webinar. 1) Data Science Salon Miami: Sidd Seethepalli's giving a talk on advanced AI systems in production. Nico, Vargas and Aaron are there too. 2) AI Hot 100 New York: Ben and I are meeting other AI native companies in the city 3) Vellum webinar: Noa Flaherty is hosting our first webinar series (FREE) with Pratik Bhat and Lior Solomon from Drata. We're sharing best practices on AI development in this series, grab your spot here: https://lnkd.in/ddc_jFBi We're out there to share our learnings in building AI systems. Meet us at any of the events or join the webinar!
-
?? Last Chance to Sign Up! ?? Today’s the day! Join us for as we kick off our Championing GenAI webinar series with Drata's AI experts, Lior Solomon and Pratik Bhat. Let's turn your best AI ideas into reliable prototypes! ??? Whether you're an aspiring innovator or a seasoned pro, this is your opportunity to learn best practices and how to navigate challenges. Get ready to dive deep into proven approaches and insights that will help you bring your AI systems to life quicker. ??? When: Today ? Time: 1:00 PM ET ?? Where: Virtual (link in comments) Don’t miss out on this chance to elevate your AI projects and connect with like-minded peers. Register now (link in comments). ?? #AI #AIDevelopment #Drata #AIPrototype #ProductDevelopment
-
-
Vellum转发了
?? Announcing a FREE new resource from Vellum ?? ?? Ever felt like you're confused with all the LLM parameters (temperature, seed, function calling, logit bias, frequency penalty), you're not alone! Model providers offer very powerful additional parameters that can control model performance and give you the steer-ability to build reliable applications in production. But they're not easy to understand, and new ones keep coming out. For instance, OpenAI just released the concept of "reasoning tokens" as part of the o1 launch last week. We're here to help! We've built an LLM Parameter Guide for you to understand how to best use these additional properties when working with models. This guide is inspired by the questions our customers ask us in Slack channels so often. I advise you to bookmark this resource because the next time you’ll work on your prompt i know that you’ll need this! It's a living resource that we'll keep updating with new developments. Link to the guide in comments, I hope you find this helpful!
-
-
Vellum转发了
?? OpenAI o1 launched today, creating a new category of reasoning models. We've added support in Vellum and published an article comparing it to GPT-4o. Main takeaways ?? 1. OpenAI o1 resets the counter with far superior performance in complex reasoning tasks. The model takes a long "thinking" step (via CoT at inference) before responding with the first token. This model is a 'preview' and I expect future iterations of the OpenAI o1 models to be far better. 2. Be very careful in using OpenAI o1 model today for production. Latency is high and unpredictable (sometimes as high as 3 minutes!), it needs very specific kind of prompts, and doesn't have support for tool calling, streaming or variable temperature 3. We tested the model against GPT-4o for its capabilities on 3 different tasks, and OpenAI o1 did very well: - Math: OpenAI o1 nailed 6/10 of the hardest SAT questions, while GPT-4o got just 2/10 - Reasoning: OpenAI o1 got 12/16 correct, just edging out GPT-4o (11/16). Both struggled with analogy riddles. - Classification: OpenAI o1 outperformed GPT-4o by 12 percentage points in accuracy, and had better precision and recall 4. OpenAI has created a new contender on the optimal quality, cost & latency threshold: - Want an extremely fast, low cost model? Go with GPT-4o mini - Want the most capable and balanced model for production? Go with GPT-4o - What to solve the most complex problems and don't have latency constraints? Go with OpenAI o1 Now more than ever, you need to evaluate models on your own use cases & evaluation metrics and pick the best model for your task. I'm excited to see how foundation models keep evolving! Full link to our report in comments.
-
???? It’s prod time! Join us for Session III of our webinar series, "Championing GenAI: Bring Your AI Systems to Life, Faster," featuring Jordan Nemrow, CTO and Co-founder of Woflow and Daniel Wiener CEO and Co-founder of Autobound. Daniel's expertise in sales, leadership, and tech led him to co-found Autobound in 2019, a GenAI-powered platform for writing hyper-personalized sales messaging. Jordan has a deep background in software engineering and technology. In 2017, he co-found Woflow, setting out to build the data infrastructure needed to power seamless connections between merchants and technology. In session III, "Prod Time: Launching Your AI Smoothly", Daniel and Jordan will review and breakdown the strategies and framework to smoothly transition your LLM applications from prototype to production. By the end of the session, you'll gain a clear understanding of the entire process needed to reliably launch your AI application. Get ready to geek out on: ?? Maintaining quality in production for consistent AI performance ??? The version control system you need for smooth iterations ?? Tips for capturing user feedback in production ?? Post-production evaluation strategies Don’t miss this webinar series! Discover the essentials for building successful AI systems. Grab your seat! (link in comments) #AI #GenAI #webinar #AISystems
-
-
Welcome ?? Aaron Levin! ??
New day new ? joining Vellum! Welcome ?? Aaron Levin to the team!! Aaron joins us as our Founding Solutions Architect -- he will help our customers build their AI systems in Vellum & find ways of improving the product, docs and onboarding flows in the process. He was most recently the Head of Engineering at Fiveable and has been a long-time Vellum customer. I've known Aaron for years, he brings incredible energy and a deep passion for helping customers succeed. Excited to work with you for the coming years!
-
-
Excited about the partnership with Woflow!
"Vellum is like our vessel for all the AI code we write — it allows us to intuitively route logic based on inputs and outputs. This makes organizing our code so much easier." - Jordan Nemrow, CTO of Woflow Woflow, a trusted partner of companies like Walmart and DoorDash, has achieved 50% faster time to market on their AI features with Vellum! Woflow faced a common challenge in the AI world: how to continuously update AI features, test models rapidly, and build fault-tolerant systems without getting bogged down in infrastructure complexities. By leveraging Vellum's platform, Woflow achieved remarkable results: ? 50% faster AI development process ? Up to 20 AI updates per week without full system redeployment ? Simplified code management and improved team collaboration ? Instant error detection and resolution Michael Liu, a Data Scientist at Woflow adds: "It's also very useful that while Vellum provides a visual interface, it's code-optional - you can do work with complex code if you want, but it's also very simple if you don't. With Vellum’s visual graph, it’s so much easier to follow the logic and improve AI features" What's particularly exciting is how Vellum has empowered Woflow to focus on innovation rather than infrastructure. They're now able to involve more team members in AI development and envision a future where their entire operations team can update prompts and create tailored AI workflows for each customer. We love working with the Woflow team and excited about the partnership for the years to come! Read more about the partnership here: https://lnkd.in/gBkvnjW5
-
?Next up in our speaker highlight for Session II our webinar series, "Championing GenAI: Bring Your AI Systems to Life, Faster", is Sebastian Lozano, Senior Product Manager at Redfin. Sebi is a dynamic product leader with experience launching innovative solutions, including Ask Redfin in 2024. He also led Mastercard Data & Services' product management team abroad and was an early writer for Morning Brew, witnessing its growth firsthand. ??Session II is all about "Building Trust in Your AI System" AKA it's time to put it to the test! Join us for an in-depth look at how Sebi and the team at Redfin successfully developed their AI-powered virtual assistant, Ask Redfin, using a test-driven development approach. Get ready to geek out on: ?? The nuts and bolts of a test-driven development approach ?? Defining success: outcomes, cost efficiency, and latency ?? Combining data-driven measures with LLM as a judge ?? Pro tips for hitting the mark with your LLM apps Don’t miss out on this resourceful session! Register now (link in comments) #AI #GenAI #webinar #AISystems
-
-
Vellum转发了
I'm excited to be a part of Vellum's new webinar series next month!
I’m thrilled to announce our new webinar series starting on September 18th, Championing GenAI: Bring Your AI Systems to Life, Faster. Building and launching a real-world AI system that goes beyond the basics requires a lot of infrastructure work. The road ahead is often unclear, and the toughest challenges usually hit at three critical stages: 1. Starting out: “How do I validate my AI idea?” 2. Development: “How do I create a reliable system?” 3. In production: “How can I keep improving my system?” The AI development journey is tricky, but luckily we now have companies who’ve successfully launched their systems — and people like Lior Solomon, Pratik Bhat, Sebastian Lozano, Jordan Nemrow and Daniel Wiener who want to pass their learnings. We’ll cover the most useful strategies to align product and engineers to ship faster, the latest techniques of building reliable systems, and the tooling you need to continuously improve your systems in production. To provide the best experience for everyone, the space is limited. You can save your spot today and learn more on the link in the comments!
-