NinjaTech AI

The day has arrived! Today is the last day to apply for the AWS Generative AI Accelerator. 🤖💡 As you put the final touches on your applications, meet a few key members of the AWS Startups Gen AI Team. We're committed to supporting the future of #generativeAI innovation. 📺 Watch as Nisha Nadkarni, AWS Sr. Solution Architect; Brenna (Jensen) Shubin, AWS Sr. Account Manager; Samantha Lisowski, AWS Sr. Account Manager; and Babak Pahlavan, founder, CEO, and CPO of NinjaTech AI, share insights into this team of specialists dedicated to your growth. Apply now: https://go.aws/3Lt5Wyy

www.MyNinja.AI reached a new milestone: 722k MAUs and 55k DAUs (thnx to all the Ninja fans around the world! You make us nervous...I mean excited! 😅 ). Our AI Agents have successfully completed 1.52M+ tasks since our launch on May 21st. An observation: Right after we launched our ImageGen agent, DAUs started to move up faster too (I'll post the graphs below). The seamless interoperability between our agents (which don't use GPUs but instead use AWS AI TRN ) and external models (e.g. Bedrock's Amazon Titan, Stable Diffusion and Dall-E) seems to have hit the right customer need. If you're curious, the order of highest used agents are: ImageGen > Advisor (for writing help) > Coder > Researcher >Scheduler. The orchestration of everything in one unified UI is holding up well. We're learning in real-time that no one-model is perfect and together-is-better (validating what Matt Wood was mentioning during the AWS keynote last week). Also, we just rolled out a few new cool things: 1. Our brand new V4.0 UI: it's even more modern & optimized to focus on results mostly. Check out the "Dark" mode; it's pretty neat! 2. V1.0 of File-upload: you can now upload PDFs and text files and ask our model or any of the 24 external models we support to analyze it and ask questions about it (our own model's context length is only 8k for now, but you can fully take advantage of Gemini 1.5's sweet 2M context window along with GPT4's 128k and Sonnet 3.5's 200k!). We'll be adding more file-upload features soon. 3. For Pro users: remember, you get access to 24 external models + Deep Researcher and you only pay for the cost of each tasks (you can see how much each task costs by looking at "account activity" from the main menu). 4. We fully implemented the latest features in Neuron SDK 2.19 (batch-size 16) and we're now using 50% less TRNs despite having doubled our user based in the last 2 weeks (literary, our AWS bill be a lot less this month despite serving more users; this is weird and awesome :D). You can read more about this here: https://lnkd.in/gifM9FKY (big thnx to Gadi Hutt Kamran Khan Samantha Lisowski Brennan Demro Nafea Bshara Swami Sivasubramanian Matt Wood and their awesome teams). Check out the Pro version of the product to get access to all the features (and don't forget to give us feedback from inside the product!): https://myninja.ai/company

www.MyNinja.AI hit 450k MAUs yesterday (thank you!) and to celebrate, we just rolled out our latest agent: "Image Generator" Our latest NinjaLLM (trained using AWS AI Trainium) can now seamlessly interact with Amazon Titan, Stable Diffusion and OpenAI's DALL-E 2 & 3 to create AI images for you: One unified interface, one billing, and you just pay for costs of each task (you get lots of images this way and you can easily compare the outputs). Amazon Bedrock was especially helpful because we were able to easily integrate with 2 ImageGen models (Amazon Titan & Stable Diffusion) and they work pretty fast. Also, Amazon Titan's quality was a pleasant surprise to us (especially for work related images or natures/animals). Huge thanks to Matt Garman Matt Wood Jeff Barr Swami Sivasubramanian Dan Grossman Nafea Bshara Gadi Hutt Kamran Khan Samantha Lisowski Roslin You Danny Meurer Paul Bernard and their teams for helping us land these new agents pretty fast (I'm not sure if many people know how great Amazon Bedrock is; it's pretty awesome! aws.amazon.com/bedrock). Lastly, the video below was from a real-world use-case 😊 (see the agents interacting with each other while keeping context). P.S: Don't forget to give us feedback from inside the product's "headset" icon. We're iterating daily and pushing code to production weekly! 🤖 P.P.S: We move fast because we have an epic team of Ninjas that lead the way at lightning speed: Sam Naghshineh Arash Sadrieh Kenneth Ho Andres Portillo Divya Chirasani Peter L. Nath Pearson Kurt W. and everyone else.

www.MyNinja.ai hit 400k monthly-active-users (MAUs) last night (thank you everyone! you make us nervous and excited 😅 ) and we just rolled out a new version last night. Here are the goodies: 1. Deep Researcher 2.0: more accurate multi-hop results (this is in the Pro version). 2. More Robust Amazon Bedrock support: Login to www.MyNinja.ai, upgrade to Pro, and you get access to 20 models from Bedrock/OpenAI/Google all in one interface! 3. Login with FB: in addition to Google, Amazon, Apple, now we support login with FB. 4. More optimized mobile UI: we've evolved our design language and you'll see us further simplifying our interaction flows even though our features are growing. 5. In honor of 4th of July, we're rolling out our $0.99 Pro version promotion (you'll get $9 free credit to use Ninja Pro!): you'll have access to Deep researcher, 20 external models, faster responses and more to come. 6. Leveraging Neuron SDK 2.19: we got early access to the latest Neuron SDK from AWS AI team and it's awesome! Magically, our speed improved and we needed 30% less of our Trainium fleet (speed + throughput improved!). Huge thanks to Matt Wood Nafea Bshara Gadi Hutt Kamran Khan Tyler Osterberg Samantha Lisowski and their teams! (remember, our models don't use GPUs and they keep getting optimized with each version release of the Neuron SDK). Next week: we're rolling out our first version of Image Generation 📷 🤗

www.MyNinja.ai is powered by NinjaLLM which is fully trained & served by AWS AI chips (Trainium). Our latest NinjaLLM achieved a breakthrough in accuracy based on NQ (NaturalQuestions) and HotPotQA (multi-hop) benchmarks which are essential for our Researcher Agent. My partner and Chief Science Arash Sadrieh with our scientists Tahir Azim Tengfei Xue along with our partners at Amazon Web Services (AWS) penned a blog post explaining how we did it. If you are an AI/ML nerd like us, you'll probably enjoy reading it: https://lnkd.in/gR7bi5af The cool thing is, Arash Sadrieh and team achieved these results by spending less than $30k for all of our training passes and no GPUs were involved (if we had to use GPUs, we estimate it would've cost around $200k). The combination of NinjaLLM and the full async/sync system that my other partner and CTO Sam Naghshineh architected, make www.MyNinja.ai a task-oriented agentic (beautiful) system. Give our Deep Researcher in the Pro version of www.MyNinja.ai a try, you might like it! :D (Feel free to use the coupon Code: MyNinja10) CC: Matt Wood Swami Sivasubramanian Jeff Barr Gadi Hutt Nafea Bshara Kamran Khan Brennan Demro Danny Meurer Dan Grossman Roslin You Faraj Aalaei Laszlo Bock Kris Vulgan P.S: Amin Vahdat Ali Ghodsi you are the experts of this field; super curious to know your feedback as well 🙏

We are thrilled to announce 📢 our latest blog post with Amazon Web Services (AWS) that reveals🔍our Fine-Tuned science and technology 🦾 within Ninja! You can read the brand-new blog post below ⬇️ and try Ninja at MyNinja.ai

It took 29 days for www.MyNinja.Ai to reach 100k MAUs and then 5 days to reach 200kMAUs and then last night Ninja reached 250k MAUs: Thank you to all of our users; we are nervously excited about the accelerated growth :D Also we just rolled out our latest version of the product which includes: 1. Our latest Ninja LLM: it's 2sec faster (try it!) and it reached state-of-the-art RAG accuracy compared to other open-sourced models (based on NQ & HotPotQA benchmarks - more on this to come later). This model was trained fully using AWS TRN1 and (zero GPU is used in training & serving of Ninja LLMs). 2. We now have 20 external models seamlessly enabled inside the product including 14 models from Amazon Bedrock (includes Sonnet 3.5!) + OpenAI GPT 4o and Gemini 1.5 Pro. You can use these models to compare & choose best answers for a variety of tasks such as coding, writing help, learning, explaining hard topics, language translations, etc. 3. We now support multi-login via Amazon, Apple, Google and email (FB and Microsoft will come soon). 4. We now have a faster & more optimized UI for desktop & mobile web. By next week, we'll turn on "Image Generation" agent as well! To get the best experience, try out our Pro version. Use the coupon code (MyNinja10) to get 1 month free: www.MyNinja.ai CC: Matt Garman Matt Wood Swami Sivasubramanian Nafea Bshara Jeff Barr Paul Bernard Danny Meurer Gadi Hutt Kamran Khan Samantha Lisowski Laszlo Bock Paul Muret Faraj Aalaei Lorenzo Thione Kris Vulgan Nader Mousavi Todd Stavish Aydin Senkut P.S: our team almost pulled an all-nighter to make this launch happen Sam Naghshineh Arash Sadrieh Divya Chirasani Kurt W. Armand S. Andres Portillo Kenneth Ho and others (thank you!!!!!!)

Knowledge workers spend too much time on administrative tasks 📋, and studies by Goldman Sachs prove this 📊. 15 hours a week are spent on mundane admin tasks like meeting coordinations. For managers you spend even more time. Schedule meetings, effortlessly 🗓️ with Ninja's Scheduler agent. Ninja coordinates your meetings across time zones 🌐, ensuring optimal timing for one-on-ones and for group sessions with Pro🤝. The AI scheduler Agent does the hard work 💪, so you don't have to. Get your Ninja Today at MyNinja.ai

Yesterday marked a big day for us at www.MyNinja.AI 29 days after launch, we achieved 100,000 monthly-active-users (MAUs) with an average user session time of 9:28 min and our AI agents have successfully completed 141,000 tasks...and all metrics are accelerating! (and we haven't even launched our native apps yet). We achieved this without a single use of GPUs thanks to AWS AI chips; and as a result, our core costs are 80% cheaper (compared to using H100s) so we've been able to pass the savings to customers. Next week we have a big version upgrade coming up: a brand new model (we've been training it for the past month) that's achieved state-of-the-art RAG among all open-sourced models and it's 2 second faster on average compared to our V1 model + multi-login support + massive expansion of our external model support by integrating into Amazon Bedrock + a few other cool things. This is a team sport and our progress has only been possible thanks to the genuine support of Matt Garman Nafea Bshara Matt Wood Swami Sivasubramanian Jeff Barr Paul Bernard Dan Grossman Jason Guesman Gadi Hutt Danny Meurer and guidance of Paul Muret Laszlo Bock Faraj Aalaei Nader Mousavi Todd Stavish Kris Vulgan and many others. Me & my partners Sam Naghshineh Arash Sadrieh and rest of our Ninjas are crazy excited (and a little bit nervous) as the rocket ship takes off 😅 (Kurt W. keeps showing me expansion plans while telling me to breathe :D). Feel free to use "MyNinja10" to get 1 month free to sign up with Ninja Pro; you don't want to miss the upgrades that are landing next week : www.MyNinja.AI CC: NinjaTech AI Amazon Web Services (AWS) AWS Startups AWS AI

Thank you AWS Startups 😅🤗🥷 www.MyNinja.Ai