rajesh pandey (@RPandeyViews) profile picture

Principal Engineer #AWS #Lambda @awscloud | Formerly Alexa AI. Love distributed systems. Chess player. Disclaimer: Opinions expressed here are 100% mine.

Seattle, WA
Joined March 2010
Photo by rajesh pandey (@RPandeyViews)Photo by rajesh pandey (@RPandeyViews)Photo by rajesh pandey (@RPandeyViews)Photo by rajesh pandey (@RPandeyViews)Photo by rajesh pandey (@RPandeyViews)Photo by rajesh pandey (@RPandeyViews)Photo by rajesh pandey (@RPandeyViews)Photo by rajesh pandey (@RPandeyViews)Photo by rajesh pandey (@RPandeyViews)Photo by rajesh pandey (@RPandeyViews)
rajesh pandey retweeted
"AWS Lambda now supports a 30-minute timeout for on-demand Event Functions". AWS Lambda now supports a 30-minute function timeout for Event Functions that run on demand, doubling the 15-minute limit for synchronous, asynchronous, and ESM invocations. aws.amazon.com/de/about-aws/…
2
11
769
Telling everyone!
YOU CAN PUT YOUR LLM IN A LAMBDA FUNCTION! TELL EVERYONE! LAMBDA FUNCTIONS CAN NOW RUN ON GPUS!
68
rajesh pandey retweeted
Another big release from the Lambda team at AWS: Lambda Web Functions! Setting up an Express or Next.js app on AWS Lambda has meant wrapping it in an adapter, paying while it waits on a database, and setting up all the multi-region routing yourself. Many teams end up using containers instead. I got to try this ahead of launch and have been waiting patiently for it to go public. Lambda Web Functions runs your port-listening server as is. A single environment can handle many concurrent requests, and billing is based on active CPU, which is typically quite low while waiting on I/O. You still pay for memory while the environment is up, so it's not free when idle. Check out this launch post from Micah Walter, with some tips to get you started. lckhd.eu/eiD5jL #AWSLambda #Serverless #NodeJS #AWS
4
13
91
6,030
Except ASE (IDK; ChatGPT it). Used other modes long. Vid gen still robotic, so did VOs & better scripts etc. to humanize. Not there yet, but will get there I feel.
We'll be spending a lot more time trying to understand the outputs of language models. A few thoughts, tips & tricks: Writing. Something I've had success with: Ask your LLM to explain something in ASD-STE100, it's a controlled language specification originally developed for aerospace maintenance documentation. LLMs well-versed in this language and it comes with heavy constraints on clean writing style that I often find a lot more readable. Sometimes I've tried to soften it a bit e.g. ask for "80% of the way to ASD-STE100" because the spec is quite stringent. But even better: Diagrams / images. Instead of writing, ask your LLM to create a diagram. These can be a lot easier to process, parse, and understand. But even better: Web pages. Ask for output "in HTML" to get a beautiful, interactive webpage. LLMs are getting really good at frontend and can create beautiful experiences, animations, etc. But even better: Explainer videos. The output format I am most bullish on is fully custom / bespoke explainer videos generated on any arbitrary topic. Experiment with things like "Create a 3b1b style video explainer on X. Use my ElevenLabs API key for audio narration". (you'd need an API key for the latter or you can ask your LLM to find you decent free alternatives that use your local compute). This is actually starting to work! In summary: - As LLMs get better, they will do more and more of the legwork autonomously, and a lot more of our work will rise up the abstractions into oversight and understanding. - Luckily, LLMs can help here too because as intelligence and code are increasingly abundant, you can ask for large, custom, discardable software artifacts (e.g. web apps, video explainers) that would have never made sense to create before. Push the boundaries here and you'll be surprised.
26
rajesh pandey retweeted
New custom event buses in Amazon EventBridge just launched!! I've written about EventBridge before and called it my favorite AWS service. I got to try these new custom buses over the last few months and I'm really looking forward to seeing what everyone builds with them. The default event bus and all its great functionality remain, as do existing custom buses (now labelled "classic"). The new, more powerful custom bus is an option you can adopt at your own pace. It can be shared across an organization through AWS RAM, a new Subscriber resource bundles the filter, target, retries, and DLQ, and event ordering is supported. Check out the launch blog from Micah Walter. Congrats to the EventBridge team! lckhd.eu/yQ1aOV #AmazonEventBridge #Serverless #EventDrivenArchitecture #AWS
2
3
1
21
2,099
rajesh pandey retweeted
Today we're excited to share Strands Harness. Strands Harness makes it easier to build reliable, cost-effective, agents with any model. It's open source, too! And equal or better performance to proprietary harnesses with 28% fewer tokens!
15
34
7
228
33,918
Additionally, you can now run 90-minute functions on LMIs.
I recently wrote about Lambda Managed Instances. Traditional Lambda allows a single request per execution environment, so a 150 MB catalog loaded into memory by a 100 concurrent requests loads a 100 times, with a potential cold start and a bill for each. The article includes a product similarity engine built on LMI, set up with my favourite Infrastructure as Code tool, Terraform: about 200 lines across four modules. The break-even for LMI vs standard Lambda is around 2.5M requests a month @ 4 GB and 200ms. LMI does not automatically scale to zero. In my tests two instances kept running overnight with no traffic at all, which is where the $2 to $4 a day comes from. AWS has since added scheduled scaling, which lets you drop that floor on a cron. lckhd.eu/Ama0oE #AWSLambda #Terraform #Serverless #FinOps #AWS
45
rajesh pandey retweeted
"AWS Lambda durable functions integrates with Pydantic AI". AWS Lambda durable functions saves your Pydantic AI agent's progress as it runs, so after an interruption like a timeout, your agent resumes from the last completed step instead of starting over. aws.amazon.com/de/about-aws/…
1
1
8
1,436
rajesh pandey retweeted
.@CollinsworthPFF knew a big @jaxon_smith1 play was just around the corner.
52
139
7
2,642
566,033
rajesh pandey retweeted
I just published a manifesto for all the developers out there who use an AI coding tool, but feel like they're not shipping much faster I wrote 10 principles for changing how you build software with AI, based on what I've seen work for teams across Amazon kiro.dev/topics/frontier-eng…
52
252
33
1,894
412,302
rajesh pandey retweeted
"Announcing 90-minute function timeout on AWS Lambda Managed Instances". AWS Lambda now supports a 90-minute function timeout for asynchronous and event source mapping (ESM) invocations on AWS Lambda Managed Instances (LMI), a capability of AWS Lambda. aws.amazon.com/de/blogs/comp…
1
2
1,325
rajesh pandey retweeted
Since launch I've been spending time with Kiro Crew, and its memory system is what holds my attention. What it keeps, what it compresses, what it lets go. Our engineers started from pure engineering constraints and landed where evolution did long ago: the brain. allthingsdistributed.com/202…
Today we're launching one of the most exciting ground up innovations in the agent landscape: Kiro Crew! What started as a side project inside @amazon (known as Meshclaw) grew into something tens of thousands of builders across the company relied on daily. It's now open source for everyone. @kirodotdev Crew is a persistent, self-evolving workspace that keeps work moving, online or off. It's built for engineering work that goes beyond a single chat session, and spans repos, tools, and days. Kiro Crew remembers your project context and history across sessions, turns your corrections into durable lessons and reusable skills, and keeps recurring jobs running on schedules you set. Devs can run several efforts in parallel or hand work to subagents that report back, so nothing waits in line. Hundreds of contributors shared feedback with us, helping make this launch possible. See how you can build your crew: kiro.dev/crew/
7
14
1
106
38,392
rajesh pandey retweeted
Monumental accomplishment!
We’re sharing a solution to the Navier-Stokes Millennium Prize Problem, one of the deepest problems at the frontier of mathematics. The proof was produced by a group of agents, using an OpenAI next-generation model significantly more capable than GPT-6 Astra. The problem concerns whether the description of smooth three-dimensional fluid motion modeled by the Navier-Stokes equations can break down. It has remained unresolved for roughly 90 years.
30
37
2
1,186
86,735
rajesh pandey retweeted
Out of curiosity, how does one build something that coordinates 10K AI agents to solve a single problem?
327
37
31
1,629
298,051
rajesh pandey retweeted
Teams building agent tooling have to decide where the code their agents write actually gets run. Spinning up a dedicated EC2 instance per agent task is not practical at any real volume, but isolation between tasks is usually a firm requirement. Lambda MicroVMs is one option worth a look. A Firecracker VM per session, session state kept for up to 8 hours across suspend and resume, and 2GB of baseline memory that bursts to 8GB when a session needs it. Suspended sessions only cost you the storage of the saved state and no compute. Elias Virtanen shows the full setup below, from the IAM role split to idle policy tuning. lckhd.eu/11wPhR #AWSLambda #Serverless #Firecracker #CloudSecurity
2
3
1
13
699
Totally digging this pun trend. Without Llama, main Meta nahi soch pata 😂
this is a banger. clearly I need to spend more time on Indian / desi twitter good thing mera calendar OpenAI
1
122
Haha, this is actually refreshingly funny 😆
Call me Andrej coz without Claude mai kaam nahi Karpathy
94