7/7 🎙️ So, if you are building with LLMs on AWS, or trying to turn a promising prototype into something production-ready, I think you’ll enjoy this one:
awsbites.com/153-llm-infe...
Would love to hear what Bedrock surprises you have run into so far 👀
#AWS #AmazonBedrock #LLM #AI
6/7 🧱 One of my favourite parts of the episode is the discussion around structured outputs.
Because in many real applications, the challenge is not “can the model answer?”
It is “can the model answer in a format my system can reliably use?”
5/7 ☁️ Bedrock is interesting because it gives you a more unified way to work with models inside AWS, with things like IAM, monitoring, auditing, and private networking options.
That makes a big difference when you start thinking beyond a prototype.
4/7 ⚠️ There is a lot of AI content that stops at “look, it works”.
But once you want to ship something, the questions change very quickly:
Can we trust it?
Can we keep it under control?
Can we integrate it safely?
Can we make it robust?
3/7 🔍 We get into:
• what inference actually means
• how to think about model selection
• how token pricing affects cost
• quotas and throttling surprises
• IAM and access gotchas
• why structured outputs are so important
2/7 🧠 We have been building AI-powered applications on AWS for ourselves and with customers, and along the way we picked up a decent collection of battle scars.
So this episode is not about hype.
It is about the practical stuff that shows up when you try to build something real.
AWS Bites Podcast cover art showing a bed in a grassy field with a huge rock resting on top of it, under the title ‘getting started with Amazon Bedrock.’ Eoin appears on the left behind the rock, shading his eyes, while Luciano stands on the right. The scene is warm, playful, and photorealistic.
1/7 🚀 Getting an LLM demo to work is not the hard part anymore.
The hard part is making it reliable, predictable, and affordable in production.
That’s what @eoinshanaghy and I focus on in the latest AWS Bites episode, all about LLM inference with Amazon Bedrock 👇
Very good point. I'll add that to the article! Thanks a million
Link: nodejsdesignpatterns.com/blog/nodejs-...
If you work with file paths and user input in Node.js, this one's for you.
I wrote a guide on preventing path traversal in Node.js:
✓ Why path.join() won't save you
✓ Attack vectors (double encoding, null bytes, symlinks)
✓ A safeResolve() function with detailed explanation
✓ Lots of code examples
✓ How to test it
Here's the thing about becoming a senior engineer:
It's not about writing clever code.
It's about knowing what can go wrong.
Security, edge cases, failure modes. The stuff that doesn't show up in most tutorials.
Path traversal has been one of the most exploited vulnerabilities in recent years.
Apache, Rails, popular npm packages, even Node.js itself have all been bitten by it.
Yet it's still way too easy to write vulnerable code.
The developer who wrote it had years of experience.
They just never had to think about path traversal before.
This wasn't a junior mistake. It's a knowledge gap.
Database credentials. SSH keys. Application secrets.
All accessible with a simple HTTP request.
If that doesn't sound scary, I don't know what does.
The code joined a user-provided filename with an uploads directory and served the file.
Looks safe, right?
But the input `../../etc/passwd` sailed right through `path.join()` and exposed everything.
I almost approved a security hole last year.
A simple image server. Clean code. Worked perfectly.
It also let anyone read any file on our filesystem.
Here's what happened 🧵
Thank you very much!
This conversation lead me to do a bit more research and I think there's still value in using http/https or undici directly when performance matter.
I have just updated the article to reflect these findings! :)
Thanks once again for providing feedback!
I thought they had resolved a good chunk of the performance issues from the initial implementation. Is there any official documentation or benchmark that tracks this?
Also, isn't undici the default fetch implementation in Node.js?
Feedback welcome.
What’s your default HTTP client in 2026: built-in `fetch()`, `http`/`https`, axios, got, ry, or something else?
#nodejs #javascript #backend #webdev #testing
If you’re writing Node services, CLIs, scrapers, or anything that talks to APIs, this should save you some pain 👇
nodejsdesignpatterns.com/blog/nodejs-...
And also:
• safe query params (goodbye string concatenation)
• streaming requests (uploads) and streaming responses (downloads)
• mocked requests with undici’s MockAgent + `node:test`
More:
• reading JSON, text, bytes, and streams properly
• timeouts (no more forever-pending requests)
• retries + backoff
What it covers:
• `fetch()` vs `http`/`https` (and when you might still need the latter)
• why `await fetch(url)` returns a `Response`, not the body
It’s beginner-friendly.
No assumptions about the tricky bits.
I call out what can bite you and what you can do about it.
That’s why I wrote a detailed blog post about it.
A practical guide to using `fetch()` in Node.js without getting surprised in production.
Cartoon illustration in flat colors: a smiling developer in a blue hoodie points upward while holding a green “Node.js FETCH GUIDE” book. Above him, speech bubbles read “npm axios”, “http/https”, and “fetch()” with a question mark. On the right, a cool turtle wearing sunglasses rides a rocket labeled “V8”; a checklist and stopwatch sit on the desk against a soft, patterned sky background.
Confession: my fingers still type `npm i axios` on autopilot from time to time 😅
...And yet… Node.js has shipped a built-in `fetch()` since v18 (2022)!
Still, I keep running into the same questions:
- Should we use it?
- When is it a good fit?
- How do we avoid the usual footguns?
Thrilled to announce chapter 09 of the Rust Lambda book.
In this one, we cover all things observability. If you're struggling to really understand your serverless applications, this one's the one for you.
@loige.co 👀
#serverless #rust #rustlang
rust-lambda.com
cool idea, but i see zero entries right now :/ did a last minute deployment go wrong?
I filled out the 2026 #AWS @Answers4AWS survey. You should too! answersforaws.com/survey/ #aws
AWS Lambda just got less #serverless 🚨
@loige.co and @eoin.sh explore Lambda Managed Instances - #Lambda functions running on #EC2
We built a video processing app to test it. The verdict? No more cold starts (kinda), concurrent invocations, and a new scaling model!
➡️ youtu.be/i-Nr0cqrdDk?...
