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 (499)
Welcome everyone to dev.to! Glad you are here and hope you are well! I am a DEV Community Moderator and my primary goal is to support everyone on dev.to and ensure everyone is having a good time!
To get started,
Get Started on Dev.to! A Beginner's Guide to Engage with the Community! 💡
Any questions about DEV and want to get answers from a DEV Community Mod? Leave a comment and come chat here!
Insights on Sloan and the MLH acquisition
Ask a DEV Community Mod! 🚀
Make sure to check out other resources here by other community members on dev.to: dev.to/help/community-resources
Feel free to introduce yourself and welcome others by replying to at least 2 people! It would be greatly appreciated! :D
Hope you are having a great summer and enjoy your stay here on DEV :D
👋🏻
😍
hi 👋🏻
👋 Hi, there!
Welcome everyone to dev.to!!! This is @dj29, I've had a great experience on this community in this short period.
I'm also a trusted member on this platform.
I recently wrote a post on how DEV COMMUNITY helped me get my 1st job as a fresher, check it out and you can ask questions in the replies anywhere, just tag @dj29, I'll answer it.
How I Got a Job Offer as a Fresher in an AI-Fearing Market — Without LeetCode or Club Leaderships on My Resume
Great!
Thanks!
Great Bro!
Thanks Bro!
오 좋은 소식이네요. 일자리를 받는 방법을 알려주시니 감사하네요.
તમારું સ્વાગત છે! હવે મહેનત ચાલુ રાખજો, નોકરી તો મળી જ જશે. શુભેચ્છાઓ! 😄
Nice that's great !
Thanks @rajneesh0021!!
Hey DEV community! 👋
Software Engineer here (@tecleadev), mostly working with C# / .NET, Docker, and Linux. I joined to exchange experiences on software architecture, DevOps pipelines, and keep up with what's happening across the ecosystem.
Glad to connect with fellow devs around the world! 🚀
¡Hola a toda la comunidad de DEV! 👋
Por aquí un Ingeniero de Software (@tecleadev), trabajando principalmente con C# / .NET, Docker y Linux. Me sumo para intercambiar experiencias sobre arquitectura de software, pipelines de DevOps y estar al tanto de lo que pasa en el ecosistema.
¡Un gusto conectar con desarrolladores de todo el mundo! ☕💻
Welcome @tecleadev,
I've been recently placed to a .NET role from college placements, would love to learn from you.
👋 Hi everyone! I'm a SaaS developer focused on virtual card issuing APIs, batch card issuance, and payment integration. New to Dev.to and excited to share practical guides on API integration, SaaS architecture & compliance. Always happy to discuss virtual card issuing / API needs with developers and businesses. Great to be here!
Hola!
Welcome aboard! Awesome tech stack—C#, Docker, and Linux is a solid combo. Happy to connect and exchange architecture ideas!
Hi everyone, I'm Daniel.
Freelance, TypeScript, lots of agent PRs.
I am building Lumpcode, a git-first CLI for long agent campaigns: one context, one branch, one PR, resumable from remote git. Early development, Apache 2.0.
Here for how-tos and for people who are already looping agents instead of chatting them.
github.com/lumpcode/lumpcode
Resumable from the remote means the run survives the process dying. That's the failure I keep hitting: a long run stops halfway and whatever only lived inside the harness is gone.
I got there from the other side on a multi-agent game. No provider session APIs, every call resends the whole history, which sounds wasteful and is the best part, because the history is mine to read or rewrite. Each step's state is saved too, so a failed step just reruns, or I point a different model at it and rerun that one step.
On resume, does the agent get its earlier reasoning back, or only the repo state?
You have 2 ways to resume your cycle depending on your needs :
The unit of work for a lump campaign is a 'context', contexts are intended to not share the same sessions for their runs, but you can use artifacts created from previous contexts for other ones (.md files for example). I use this kind of pattern for my abstraction example. One lump create a md file, that you review and that will be used by another context. This is using the state of the repo to agrement your next context state. Here is the article for the abstraction lumps : dev.to/dyod/turn-a-dupes-report-in...
Resumability of the agent session itself. The lump runs can keep their history and with it the whole session id for your corresponding cli. You can easily reuse a session id for a run (even automatically in the loop config by reading if a previous history exists), and take the same session. Nice for error resumability, or tweaking what an agent has done before merging. I do not recommend this for multi contexts sharing. History is gitignored by default, but nothing prevents you from commiting them if you want, files are isolated by contexts so there will be no conflicts.
The marker commits and the .md handoff both sit in the repo, so a fresh clone on another machine picks up where the last run stopped. That one I get.
The session id is different. Reusing it asks the CLI to find its own transcript, and that lives in the home dir of the machine that ran it, not in the checkout. Commit the history files and another box gets the id without the conversation it points at.
Is that session id only good on the machine that made it, or does Lumpcode keep enough of the run to rebuild it elsewhere?
The session ids by themselves can only be taken back on the machine with the corresponding agent CLI.
But today, my workflow basically is like this without needing a fully recoverable transcript :
If you want a full transcript of what happened on the branch, the history could be used as a commitable transcript but today it is not exactly the intention. The history is only created at the root of the workspace (and not in the worktrees of each lump context), so it will not be commited reliably without some tweaking. I have on my todo an history cleaning because I am not satisfied with it, but with your insights it gives me some ideas to make it easily commitable and usable as a nice transcript too.
Welcome to the community, Daniel! Looping agents with a git-first approach solves a huge pain point for long runs. Excited to see Lumpcode evolve!
Hey! I'm Francis. I've been building software for a while, and lately I'm working mostly solo using Claude Code as a daily collaborator rather than just autocomplete. Just shipped a hosted changelog tool called Chngd and posted my first write-up here about the AI drafting pipeline and a Supabase RLS mistake I made along the way. Glad to be here.
Read your writeup. PostgREST sits in front of every table whatever your own server code does, so "I don't use the client library" never meant the table was protected, and the anon key is public on purpose.
You don't need a special security model to find that kind of thing. Point any decent model at your own schema and settings and it turns up whatever got left on defaults. Holes like this are everywhere and we still haven't got used to how easy they are to find now.
Did the alert name the exposure, or did you have to go looking?
👋 Hi everyone! I'm a SaaS developer focused on virtual card issuing APIs, batch card issuance, and payment integration. New to Dev.to and excited to share practical guides on API integration, SaaS architecture & compliance. Always happy to discuss virtual card issuing / API needs with developers and businesses. Great to be here!
Hi, I'm Billy. Eighteen years building software, most recently as CTO of a SaaS startup.
I spend a lot of my time training small language models on a single consumer GPU, and what I like writing about is not the training but the measuring, including the times the measurement contradicts me. Which is often.
The current one plays chess. 142 million parameters, trained from scratch, no fine-tuning. It plays rated games on Lichess and it will take a challenge from anyone: lichess.org/@/philidor-142M
I'll be posting about what it taught me, mostly around evaluation and how easy it is to fool yourself with your own benchmark.
Glad to be here.
Went and read the chess writeup. 98.85% legal sounds high, but one illegal move loses the game, so the masked softmax over legal moves is doing the real work.
I run a long social deduction game where LLMs have to follow rules over a growing event log. I can't touch the sampler on a hosted model, so I do the same thing in the prompt: at each step I paste in only the rules that matter right then, as the last message, and check the answer against the list of allowed options before it counts. That catches a bad answer instead of blocking it.
When the top pick is illegal, is the next legal one usually still decent, or is the model already lost in that position?
Hey everyone! 👋
After ~10 years directing live gaming products and large-scale multiplayer servers (ARK, Fortnite Creative, Minecraft), I'm currently focusing on multi-agent AI systems, enforcement architecture, and LLM tooling in Python.
Recently open-sourced
loopwardto enforce structural agent boundaries. Always happy to chat about multi-agent orchestration, mechanical constraints, or game dev background!Hey everyone — Bryan from North Carolina. Self-taught since April of this year: started out copy-pasting code from a browser chat with zero understanding, and somewhere along the way it turned into building a live AI receptionist business and a whole accountability harness around my AI (it is not allowed to tell me something works without proof — long story, it's on my profile). Day job by day, building in the margins at night. Joined this week, already smashed two bugs into the Bug Smash challenge. Happy to trade notes with anyone self-taught or anyone trying to make AI agents actually trustworthy — those are my two favorite conversations.
that sounds awesome and I am on the similar journey of hustling on my own during nights.
I honestly never knew I would become so enthralled by every aspect of this. Good luck on your journey, and have a wonderful day!
yeah same same, I am not a developer and I love building it now on my own. BUT, I know that if the startup should be succesfull, there will come a day when some serious dev will take over building it.
The gap between a developer and a determined explorer is closing rapidly. I don't consider myself a developer either, but I’ve learned that if you stay with it, you can accomplish far more than you initially thought possible.
I’m an introvert, and it was genuinely difficult for me to start posting or even showing people my work. But the truth is, we’re often our own harshest critics, and every piece of honest feedback helps us grow. I don’t think you need a “serious dev” to swoop in and take over what you’re building. Keep learning, keep building, and don’t count yourself out. The serious developer you imagine needing someday may be the person you’re becoming right now. its all a state of mind
Yes thats true. However I am actually building developers tool - AI powered NSFW content moderation API and dashboard.
Right now its quite basic as I have only 1 LLM model checking the content, but I definitely need to upgrade to be able to detect multiple categories, able to fine-tune the models, etc. I can do it on my own, I bet that, but the results for devs might be inaccurate. Will see where the future goes :)
hey guys im uh a bit new to this but i have dabbled with coding in the past i tried to make an app in xcode with chatgpts help of course called ramp finder if you want more details on that id be happy to answer those questions i did take an html class in college but had to drop it right now im taking codeacademys front end engineer career path wish me luck
Welcome to DEV Jj!
im kinda intimidated by the unit sizes im afraid if i bit off more than i can chew
I always lived by the idea that if you are uncomfortable but not dead, you are getting stronger. I bet you will find you learned quite a lot through the process. Just take it one step at a time in bite sized chunks and make sure to give yourself breaks!
Hello and welcome, Jj! creating apps using AI is a very common thing though just a piece of advice: projects never finished. Always plan your next move and improve your project constantly. Good luck!
In the past projects never finished because time. Now never finish because always you can do something else!
Hey DEV 👋 I’m Drew.
I somehow turned “I wonder what’s inside a Fortnite replay file?” into a 🐇 rabbit hole involving game data 🎮, AI agents 🤖, UEFN 🛠️, and now this account. 😂
I’m here to build weird useful things, learn in public 🧠, share what works (and what catches fire 🔥), and meet people doing the same.
What are you building right now? 👀
Replay files are the good kind of rabbit hole. No spec, undocumented binary, and the only thing that knows the format is the game writing it.
Building a social deduction game where LLM bots play against people right now. 24 models across 11 providers, all interchangeable. Most of the work went into keeping a bot on the rules over a long game.
What did the replay format turn out to be good for?
Some comments may only be visible to logged-in visitors. Sign in to view all comments.