Leave a comment below to introduce yourself! You can talk about what brought you here, what you're learning, or just a fun fact about
yourself.Reply to someone's comment, either with a question or just a hello. 👋
Come back next week to greet our new members so you can one day earn our Warm Welcome Badge!
Top comments (929)
Hey everyone. Hope you guys are doing well!
As mention in the last welcome thread, I am waiting to fully introduce myself on "Welcome Thread v400". @javz 4 more weeks to go :D
With that said, welcome everyone to DEV. What brings you here today?
Hello
hiii
hey
hey
hiiiii
Hello
hi Nice to meet U.
hola chaval
hi, whats up?
hey
hey
It’s getting close!
hello
Hi Lakshay, how's it going?
Not much.. You?
Been doing well!
That's great ^^
hello
hello
HI @francistrdev
hello
Hi everyone! 👋
I'm Dageno AI
Happy to be part of DEV! 🚀
hi
hello
Hello..
hii
hi
hello
hello
厉害
Hello
Hey Hey! Here to learn new experiences
Hi~~
Hi~~ is not defined
hi
hello
👋
What projects are you working on?
Hello
halo
Hello
hlo
hello
Hello
hello to the welcome thread
Hey Everyone I am a IT student from India I have attended 15+ hackthons and also am a AI-ML focus engineer looking foreward to more advanced parts of IT. Any tips or recommendations for me ?
15 is a lot
Hi nice to meet you
Hi
hello
Hellow
Hello Francis
Welcome
hello
hi
HI
Hi
Hey hoi hallo
Hello!!
Hi there!
hello
Hi
good
Hello
hello
hi
Hello!
You
Hello!
Hello
hello
hello
Hello
hi
Hi 😁
hello
Hi everyone! 👋
I'm Raza from Pakistan. I'm learning web development by building small, useful web tools.
I recently built a free browser-based image compressor for JPG, PNG, and WebP images. I'm here to learn from other developers, share what I build, and improve my projects through community feedback.
Happy to be part of DEV! 🚀
Great to be here! I’m interested in fintech, financial data, web development and practical tools that make complex information easier to understand. Looking forward to learning from the community and sharing some of my own projects and experiences as well.
Hi everyone! 👋
I’m a security-focused engineer and researcher from Japan.
I’m currently working on incident analysis, defensive security, AI-assisted engineering workflows, and technical documentation. A lot of my work is about turning messy real-world evidence into something reproducible, reviewable, and useful for other practitioners.
I’ve also started publishing some of that work on DEV and on my site, OSIIX.com.
Recently, I’ve been writing about topics such as cyber recovery, security engineering, AI workflow reliability, and lessons learned from real incident analysis.
Looking forward to learning from the community and exchanging ideas with people working in security, infrastructure, AI engineering, and reliability.
Nice to meet you all!
hello..
hello
hi
Welcome!
Hello
hii
Welcome to any news members!
Hello DEV! 👋
First post here. I build web products, and lately I’m mostly writing Go: a modular monolith with chi, pgx and hand-written SQL, no ORM. On the frontend it’s React 19, Vite and TypeScript.
My side project is an app that turns messy project input (slides, docs, meeting transcripts) into a structured discovery report: requirements, architecture decisions, milestones. The fun engineering problems so far:
treating everything the user uploads as untrusted input to the model
optimistic concurrency with If-Match on every edit
a change history you can actually restore from
I plan to write about those here. Happy to swap notes with anyone doing Go + Postgres or building with LLMs. What are you working on?
Welcome! The "everything the user uploads is untrusted input to the model" point really resonates. I hit a smaller version of it building the chat assistant on my own site: the model is only allowed to link to an allowlist of URLs from the site, so a bad prompt can't make it send people anywhere unexpected. And in a side project (an AI tutorial generator) I validate and repair the model's JSON locally, and never cache a degraded result.
Curious how you're separating instructions from document content once a transcript or slide deck goes into the prompt. Are you wrapping it in delimiters, running a separate extraction pass, or something else?
Me: mostly React, Next.js and Node.js, plus small AI experiments on the side. Looking forward to your posts on If-Match and restorable history, too!
Thanks, @sanjivsutar ! The URL allowlist is a nice touch. It limits what a bad prompt can do instead of trying to stop every bad prompt.
To your question: delimiters, but I don’t rely on them alone. All untrusted text (files, transcripts, project content) goes inside tags, with any closing tags in it defanged. Every model call also adds a system note that tagged content is data, not instructions.
The real protection is what the model is allowed to do. It has no tools, and its output is only parsed as JSON proposals like “add requirement” or “update goal”. Every record id it names is checked against the DB, scoped to the user’s own project. Removals are capped per reply and need confirmation, and nothing is saved until the user reviews it.
I don’t do a separate extraction pass, because it would read the same untrusted text. So even a “convinced” model can only make suggestions that get validated and reviewed.
More in a post soon. What’s your tutorial generator built on?
Love the "model can only propose" approach. Capping removals per reply is a smart touch.
TopicTutor is Next.js 16 on Netlify's free plan, Gemini first with Groq as fallback. One model call per topic and level, cached in Netlify Blobs for 30 days. Broken JSON gets repaired locally with jsonrepair plus a salvager, and degraded results are never cached.
Looking forward to your post!
will check and update
hi
Hi everyone 👋 I'm Hiroki (@nomunomu0504), a developer from Japan.
I'm building Forecall (forecall.dev): a small open-source linter that scores the tool descriptions of MCP servers — the text an AI agent reads when it decides which tool to call. It started as a study of 46 public servers on Smithery and grew into
npx forecall lintand a site.What I'm learning right now: a lint score tells you what's written, not what a model will do. I measured it this week and had to walk back a claim in a Chrome DevTools MCP issue. Humbling, and the best lesson so far.
Glad to be here. If you run an MCP server, I'd love to hear what you'd want a linter to catch.
Hi everyone! 👋 I'm a Python/AI developer and the CEO of a small startup in Korea.
What brought me here: I've started running LLM agents entirely on local models on an NVIDIA DGX Spark (vLLM + llama.cpp, Qwen and Gemma), and I want to share what breaks and how I fix it. My first post here is a weekly ops diary from that setup.
Currently learning: sandboxed agents and how to make small local models reliable enough for real daily jobs.
Looking forward to reading your posts!
Me too: NVIDIA DGX Spark (vLLM + llama.cpp, Qwen and Gemma) locally, it's been a lot of fun. Have needed to go cloud resource hybrid from time to time, but these local apps can carry their own weight.
sounds really cool. also im keen on visiting Korea sometime. its on my bucket list
Hey everyone! 👋
I’m an IT enthusiast from Kenya. I joined this community to learn, build, and connect with people who are exploring technology.
I’m currently learning more about web development, AI, and cybersecurity. I’m also interested in building useful products and explaining technology in a simple, practical way.
Fun fact: I can spend hours thinking about a tech idea before writing the first line of code 😂
Looking forward to learning from you all and connecting with other people who enjoy building with technology.
how's the tech scene in kenya?
Hi
welcome brother
Hey everyone! 👋 I’m a content writer and SEO specialist interested in web development, AI tools, automation, and digital publishing. I’m currently exploring new ways to combine technology with useful content and improve web workflows. Glad to be here and looking forward to learning from the community!
Hello
Some comments may only be visible to logged-in visitors. Sign in to view all comments.