Bharathan Balaji

I've never felt this much behind as a programmer. The profession is being dramatically refactored as the bits contributed by the programmer are increasingly sparse and between. I have a sense that I could be 10X more powerful if I just properly string together what has become available over the last ~year and a failure to claim the boost feels decidedly like skill issue. There's a new programmable layer of abstraction to master (in addition to the usual layers below) involving agents, subagents, their prompts, contexts, memory, modes, permissions, tools, plugins, skills, hooks, MCP, LSP, slash commands, workflows, IDE integrations, and a need to build an all-encompassing mental model for strengths and pitfalls of fundamentally stochastic, fallible, unintelligible and changing entities suddenly intermingled with what used to be good old fashioned engineering. Clearly some powerful alien tool was handed around except it comes with no manual and everyone has to figure out how to hold it and operate it, while the resulting magnitude 9 earthquake is rocking the profession. Roll up your sleeves to not fall behind.

RFT is now available on Bedrock, SageMaker, & Nova Forge 🎉 Serverless RFT, multi-turn learning, Lambda rewards, AI feedback, RLAIF + RLVR — built for reasoning-heavy agents and copilots.

Our paper on algorithms to automate a key challenge in carbon footprinting - accurately recommending emission factors across complex supply chains to enable more organizations to measure and reduce their environmental impact at scale. Link to paper: amazon.science/publications/p…

Stoked to give my first keynote at the Intelligent Environments conference yesterday on a topic that I'm proud to have been part of for the majority of career: Sustainable Buildings. The talk focused on data driven methods to get to net-zero carbon buildi…lnkd.in/gM3EXrD

Training of reinforcement learning policies can take a long time. Therefore, distributed training is key to speed up the training process. We demonstrate how you can easily use the CPUs and GPUs available in a single machine or a c…lnkd.in/gQDzYM4 lnkd.in/gyuRnJ7

Much of the online coursework we see today follows a fixed schedule. In collaboration with Stanford, we designed a personalized schedule for each student using reinforcement learning. Results published in CHI 2020. lnkd.in/gY6m-ai

Our DeepRacer paper explains how you can teach a toy car to autonomously navigate a track using reinforcement learning. The training happens in simulation and we can train successful policies in just 5 mins. The results have been published in ICRA 2020. lnkd.in/gRupvE8

If you are looking for an internship in sciences, there are a number of opportunities posted here: #amazon #science #internship lnkd.in/ga7392w

Autoscaling of servers ensures enough compute to satisfy changing demand and also reduces costs. In this blog post, we show how you can use reinforcement learning and SageMaker to proactively autoscale your fleet for gaming servers…lnkd.in/gQUMMpC lnkd.in/g-MWcu7

Reinforcement learning has become the method of choice for developing AI in games. In our blog post, we show how you can create your own AI agents for the classic snake game and use the learned agent to compete in the global Battle…lnkd.in/gtqvxMp lnkd.in/gdebgDq

AWS is looking for an ML expert to develop next generation control and automation for its data centers. In this role, you will lead multiple product development initiatives using state of the art techniques. For more details and ho…lnkd.in/geGY5GD lnkd.in/gBntDrs

With batch reinforcement learning, you can train your agent to improve decision making based on existing data instead of trial and error interactions. Our latest blog post demonstrates batch RL training using Amazon SageMaker RL wi…lnkd.in/gU5uNsy lnkd.in/gBeGDJ8

Synthetic Gestalt uses SageMaker Reinforcement Learning capabilities to accelerate their drug discovery process by 10 times! Grateful to see our customers take advantage of the tools we have developed. #sagemaker #reinforcementlearning #aws lnkd.in/gQASY2y

It was an enriching experience to participate in Amazon AI Conclave. Presented our work on contextual bandits with colleage @Saurabh Gupta and engaged with the growing ecosystem of startups in India. Excited to see these startups make up the future of Ind…lnkd.in/gUZxMub

Excited to talk about the latest in Amazon SageMaker RL with colleague Runfei (Anna) Luo and Michael Cox from NASA JPL at Re:Invent 2019 #rl #sagemaker #reinvent2019 lnkd.in/geuxhhH

At the next ACM/IEEE CPSWeek we are organizing a new workshop on Autonomy in Cyber-Physical Systems . The workshop focuses on development of autonomous CPS that can continually learn while deployed. Paper deadline: Jan 8 2020, AoE. #autonomy #cpsweek autocps.org

Benchmarks for applying reinforcement learning to resource allocation problems: #reinforcementlearning #aws #sagemaker lnkd.in/g8W84nk

Paper describing the techniques behind AWS DeepRacer: lnkd.in/gq64i8b Includes a video summary: lnkd.in/gJwJCJa #awsdeepracer