Vikram Shenoy

Senior Software Developer @ Amazon AGI

[Amazon @amazon]

.@AnthropicAI to secure up to 5 gigawatts of current and future Amazon Trainium chips and commits to spend more than $100 billion on @awscloud technologies over the next 10 years, expanding Amazon and Anthropic's strategic collaboration.

[Vikram Shenoy @vikramshenoy97]

We envision a future where customized LLMs unlock domain-specific scientific assistants, and conversational reasoning capabilities could fundamentally change how we discover new medicines. (2/2)

Read more at Amazon Science: https://t.co/IL3mWjeH1R

[Vikram Shenoy @vikramshenoy97]

AI is transforming drug discovery. Using Amazon Nova Forge, we helped @NimbusTx build a customized LLM that matches the accuracy of multiple specialized models in molecular-property prediction.

This is just the beginning. (1/2)

@AWSAI @AmazonScience

[Rohit Prasad @RohitPrasadAI]

Since launching Amazon Nova Forge, we’ve had great conversations with the community about what it can do and why we built it. I explore the technical challenges we navigated to make "open training" a reality here: https://t.co/zZembNDNoY

We built Forge initially because our own internal teams needed it. Across Amazon's diverse businesses, teams needed models with deep expertise in their specific domains – and our external customers wanted the same.

There were two key challenges to overcome. First, how can massive amounts of domain-specific data be infused into the model without catastrophic forgetting of its foundational capabilities? For example, if you have specialized data for drug discovery, can you incorporate it during early training stages like pre-training and mid-training? Second, how can we make the toolchain and recipes self-serve for customers?

This required solving a collection of science and engineering problems – from data engineering (what data, in what proportions, at what stages), learning dynamics (how to keep training predictable as the data distribution shifts), and reliable scaling on Amazon SageMaker AI.

We spent months refining these recipes with internal teams, then validated them with early customers to make each training stage accessible and predictable. Each time a customer used Forge recipes, their model significantly outperformed other leading alternatives.

This is just the beginning of what's possible when we lower the barriers to frontier model development. Thank you to the partners who worked with us from design to general availability. Excited to see what organizations build through this "open training" paradigm.

[Vikram Shenoy @vikramshenoy97]

!!!

[Connor Shorten @CShorten30]

Happy re:Invent week! Super excited to share our integration with Nova Embeddings in the Weaviate Database with multi2vec-aws! @antas_marcin 🔥

Also really excited to share our notebook testing the Nova Prompt Optimizer! Huge thanks to @vikramshenoy97 for explaining this at DSPy Boston! AWS is killing it with Nova Forge and customizing foundation models, definitely recommend checking it out! 🔥

[Connor Shorten @CShorten30]

Happy re:Invent week! Super excited to share our integration with Nova Embeddings in the Weaviate Database with multi2vec-aws! @antas_marcin 🔥

Also really excited to share our notebook testing the Nova Prompt Optimizer! Huge thanks to @vikramshenoy97 for explaining this at DSPy Boston! AWS is killing it with Nova Forge and customizing foundation models, definitely recommend checking it out! 🔥

[Erika Shorten @eshorten300]

AWS and Weaviate
1. Multimodal Search with Nova Embeddings: Build a multimodal search system using the Nova embeddings model (https://t.co/6fkVOWFGnt)

2. Nova Customization: Use the open-source Nova Prompt Optimizer to optimize a RAG system (https://t.co/thBvlqxKlG)

[Andy Jassy @ajassy]

Really enjoyed Matt’s keynote at #AWSreInvent today. So much innovation happening in @awscloud, and you could see it with the array of launches he unveiled.

So many parts of the keynote worth watching, but will point to a few:

1/ Excited about the availability of Trainium3. Trainium2 has substantial traction, is a multi-billion-dollar revenue run rate business, has 1M+ chips in production, and 100K+ companies using it as the majority of Bedrock usage today. Trainium2 has price-performance advantages over other GPU options that are compelling, and Trainium3 will deliver at least 4.4x more compute performance, 4x greater energy efficiency, and almost 4x more memory bandwidth than Trainium2: https://t.co/B6QwUZSJYv

2/ Worth double clicking on AgentCore, which has changed the security and scalability of deploying agents into production. AgentCore is a set of flexible building blocks that can be used in any combination developers want, and AWS added two more in Policy and Evaluations. AgentCore has a lot of momentum: https://t.co/rj1l71ZCU3

3/ Nova Forge is a game-changer for companies wanting to customize a frontier model with their own proprietary data. Like equipping a young person with a better knowledge foundation to keep learning, LLMs are better able to solve problems and improve if they’re trained early on with companies’ differentiated data. To do so, companies need earlier versions of the frontier model and ability to mix their own data with the model’s data. This is what Forge provides and this “open training” allows companies to develop Novellas that are their own, optimized versions of Nova they can use for their AI apps and agents. Customers have been itching for this sort of capability, and Forge is a uniquely compelling approach: https://t.co/JVonb18QcA

4/ Agents will become the primary way companies get value from AI. We have built some compelling agents for our customers in Kiro (for coding), Quick (for knowledge workers to leverage their own data, analytics, and routines), Transform (to migrate from one software source to another), and Connect (call center agents). But there are tasks customers want agents to solve more autonomously and over longer durations, and new AWS frontier agents—Kiro autonomous agent, AWS DevOps Agent, and AWS Security Agent—are exciting: https://t.co/YtMvneJryV

5/ Finally, I enjoyed Matt’s ending 25 launches in 10 mins, both because it was action-packed and represented so much useful delivery. The reason so many people cheered these launches is because even though they’re less sexy, they’re the meat and potatoes core infrastructure needs customers have—and with so much of the total cloud infrastructure running on top of AWS, these launches will make a lot of people’s lives easier and better every day: https://t.co/CdzzzFA1o0

Enjoy (and just day 1 of announcements today for AWS :-)!

[Amazon Web Services @awscloud]

Meet Amazon Nova Forge, the easiest and most cost-effective path to your own frontier models.

* Early Nova checkpoints across pre-training, mid-training, and post-training phases
* Blend proprietary data with Amazon Nova-curated training data
* Reinforcement Fine Tuning (RFT) with reward functions in your environment
* Custom content moderation settings

[NVIDIA @nvidia]

NVIDIA and @awscloud are deepening our full-stack partnership with new technology integrations spanning cloud infrastructure, interconnect technology, open models, and physical AI.

👀 Highlights from #AWSreInvent:

✅ NVIDIA NVLink Fusion to accelerate deployment of AWS Trainium4 AI chips.
✅ Expanding the portfolio with NVIDIA Blackwell Ultra and RTX PRO 6000 Blackwell Server Edition cloud instances.
✅ Integrating NVIDIA Nemotron models with Amazon Bedrock.
✅ NVIDIA Cosmos world foundation models now available as NIM microservices on Amazon EKS.

Dive into the details: https://t.co/DDcEXMuMG3

[Vikram Shenoy @vikramshenoy97]

Nova Forge is now available! Extremely proud to have contributed to this and grateful to have worked with some incredible engineers, managers and product teams on this one. Can’t wait to see what you build with Nova today!

https://t.co/asBAV7RFbx

[Vikram Shenoy @vikramshenoy97]

🚀🚀🚀

[Amazon Web Services @awscloud]

Meet Amazon Nova Forge, the easiest and most cost-effective path to your own frontier models.

* Early Nova checkpoints across pre-training, mid-training, and post-training phases
* Blend proprietary data with Amazon Nova-curated training data
* Reinforcement Fine Tuning (RFT) with reward functions in your environment
* Custom content moderation settings

[Vikram Shenoy @vikramshenoy97]

Incredibly proud to be a part of the Nova Forge project! Can’t wait to see what customers build with this 🙂 https://t.co/ePOckZQ0KA

[Rohit Prasad @RohitPrasadAI]

We've been hard at work solving a persistent industry problem: frontier models launch with impressive benchmarks, organizations test them, then they don't work for actual needs. To help bridge this gap, we're excited to announce Amazon Nova Forge – a new way to build frontier AI models that are experts in your domain.

[Saleh Soltan @SalehSoltan]

Super excited to share that we’re launching the next-gen Amazon Nova 2 models plus Amazon Nova Forge.

Nova 2 delivers top-tier reasoning, multimodal, conversational & agentic AI. Forge unlocks custom-model training with your data. #AWSreInvent

https://t.co/m9FqkPeVm5

[Vikram Shenoy @vikramshenoy97]

🚀🚀

[WIRED @WIRED]

Nova Forge lets Amazon’s customers train frontier models for different tasks—a potential breakthrough in making AI actually useful for businesses. https://t.co/YEIjH2pQcN

[WIRED @WIRED]

Nova Forge lets Amazon’s customers train frontier models for different tasks—a potential breakthrough in making AI actually useful for businesses. https://t.co/YEIjH2pQcN

[Vikram Shenoy @vikramshenoy97]

In case you missed it, here are all the amazing talks about DSPy and Amazon Nova Customization!

[Connor Shorten @CShorten30]

DSPy Boston! 🧩🍀

I am super excited to share the recording from the event! Huge thanks to everyone who attended, it was awesome to meet so many people excited about DSPy in Boston! 💚

Also massive thank you to Vikram Shenoy (@vikramshenoy97) for his talk on Nova Customization, Noah Ziems (@NoahZiems) for teaching us about the Arbor RL framework for DSPy, and of course, Omar Khattab (@lateinteraction) for an incredible overview on all things DSPy! 🔥

Really grateful for this opportunity to share what we are working on at Weaviate with the Agentic layer around the Weaviate Database alongside @eshorten300.

I am also super excited to open-source retrieve-dspy and continue our involvement with @DSPyOSS! 🧩🐶

I hope you find this useful 👇🎉

[Rohit Prasad @RohitPrasadAI]

Today's embeddings systems often require separate models for searching & retrieving docs, text, images, videos, and audio content. With Amazon Nova Multimodal Embeddings, we've built the industry’s first unified model that processes these content types together through a single system.

We evaluated the model on a broad range of benchmarks, and it delivers leading accuracy out of the box. But the real impact will be in unlocking insights from unstructured data that was previously difficult to access. Whether it's semantic search across mixed media libraries, building more sophisticated RAG applications, or creating entirely new search experiences, this will make AI more useful for real-world applications!

[AWS AI @AWSAI]

Meet #AmazonNova Multimodal Embeddings 🚀✍️🎯

Power your AI agents & applications with the industry's first unified embedding model that supports text, images, documents, video & audio in a single model. #AWS

👉 https://t.co/911KK9REkp

[Vikram Shenoy @vikramshenoy97]

Great event and fantastic set of talks! I had an amazing time show casing Amazon Nova Customization to the community

[Connor Shorten @CShorten30]

DSPy Boston is a wrap! 🧩☘️

What an awesome night! It was incredible to meet so many people excited about DSPy! The talks from @vikramshenoy97, @NoahZiems, and @lateinteraction were all super inspiring!

Hope to be apart of more of these in the future. Thank you so much to everyone who attended, it was a really cool experience! 🙏💚