👋👋👋👋
Looking back on your week -- what was something you're proud of?
All wins count -- big or small 🎉
Examples of 'wins' include:
Getting a pro...
For further actions, you may consider blocking this person and/or reporting abuse
I am proud of releasing an article on Gemma 4 featuring @codingwithjiro, @konark_13, and @javz! This took a lot of time and effort! Glad to meet them on zoom outside of DEV for the first time! If you like to take a look, feel free to read it below:
Gemma 4 Challenge: Write about Gemma 4 Submission
Should you use Gemma 4 for your Development? A Multiversal Analysis to Determine if Gemma 4 is Right for You!
Thank you guys for sharing your Gemma 4 experience!
@jess Hope you like this submission as well :D
It's an honor for me as well. Let's go DEVengers!!!
@francistrdev @javz @konark_13
Really glad to working with an awesome developer like you. Learned a lot and your part in the article is much more easy and fun to read. You write pretty well. Keep writing and building.
Was super fun collaborating with you on this challenge! Cheers to more in the future! @francistrdev @codingwithjiro @konark_13
hi dear can you spare few minutes here"dev.to/halakabir234hub/looking-for..."
It was too much fun collaborating with you all @francistrdev, @codingwithjiro and @javz. The thing I missed was a little of chaos haha as everything went pretty smooth and Francis made the article even more fun and informative with his choice or words and awesomeness.
This week was super productive for me, full of things I've done for the first time in my life! Some highlights:
Gemma 4 Challenge: Build With Gemma 4 Submission
Recycling made easy: a Polish recycling assistant powered by Gemma 4
I'm very grateful and hopeful for what the future holds 🌻
Wow, what a range of activities!
A nice list of accomplishments!
As a Notion MCP challenge winner, I had a meeting with Ivan Zhao - the founder of Notion himself, along with the awesome guys from Major League Hacking and another winner of the same challenge, whose project was super cool.
To say that it was a great experience is an understatement. I still can't comprehend somethink like that actually happened to be honest. 😄
Today I got my 32 Week Community Wellness Streak Badge 🥳
Thank you for being part of the community :)
Thank you, @jess 😍
This community has been one of my favorite places on the internet. Happy to be here and excited to keep the streak going.
how you got this?
You earn it by posting at least 2 comments a week for 32 weeks in a row
thanks!
Applied for the "Gemma 4 challenge" ; if you like french accent, open source and Kaggle, this may interest you:
Haha great pitch
LOL thanks... some may find this sexy or attractvie...;-p
That was so spontaneous 🤣
This week's win: tracked down a gnarly production race in a RabbitMQ→backend consumer. A POST that exceeded the timeout was being retried — but the backend was still processing the original request, so the retry fired a concurrent duplicate and they deadlocked each other.
The fix shipped as a declarative feature in my open-source framework, Mycel (config-driven microservices in Go): you can now say on_timeout { action = "ack" } to drop a timed-out idempotent request instead of retrying it, and the dedupe layer commits its fingerprint on that ack so the redelivered duplicate gets filtered too. Two releases, end-to-end fix, no application code.
Writing the whole thing up properly soon — but the repo's the real win this week 🙂
Revisited one of my earlier article for instruction system capability ladder
Also released Reporails 0.5.10 with included MEMORY checks and broken markdown link validator.
Busy busy week
Created a MCP and did a demo on how helpful it is to find and apply to a job :
Released new security trends about OpenBao
Collaborated on this post alongside @francistrdev , @codingwithjiro and @konark_13 !
Was a fun experience and felt refreshing to work on something as a group vs writing individually on DEV.
Also wrote my individual weekly post here and made progress on some side projects.
Been taking dance lessons and feeling progress. Rumba and Foxtrot are the dances I am focussing on atm.
Have a great weekend all!
It was awesome working with you. You are an awesome developer and even more awesome person.
We need to see your moves someday. Maybe a small performance in the Toronto tech events. Our very own dancing developer who explains concepts using dance moves. Awesome keep learning, building and dancing.
Thanks Konark for the nice words! You are also awesome!
Haha that sounds like a cool concept.
Finished up my first semester of teaching! Still working on responding to final projects and entering grades, but that will be a post for next week. 😅
What are you teaching?
"Visual Communication and Interactive Design" 😝
Basically it's an intro to html, css, Photoshop and Illustrator. I loved it!
Sounds interesting! No wonder you love it :))
I got my Relational Database Certificate today from freeCodeCamp!
That's nice
Hahaha take a look at this! Do we have 64-weeks badge here? 😅
We need a fully grown plant lol
Built and shipped two open-source MCP servers this week.
ThinChain : compresses options chain data 95% before it hits your AI agent. 482 rows down to 25.
GadsChain : lets Claude manage Google Ads directly. Started because my restaurant's ads were broken and I couldn't get AI to actually read AND fix them - only read.
Both live on PyPI and the Anthropic MCP registry.
github.com/SnipMCP/thinchain
github.com/SnipMCP/gadschain
This is what SnipMCP builds domain-aware AI connectors that don't just read your data, they act on it.
My wins this week — three challenge submissions, all live.
Gemma 4 Challenge → dev.to/dannwaneri/cloudflare-depre...
Migrated a production system off @cf/moonshot/kimi-k2.5 after Cloudflare announced deprecation — 22 days, 100k+ indexed documents, daily cron. Replaced it with Gemma 4 MoE as the reflection layer, benchmarked both models against 9 real queries, and wrote up everything that broke (especially the max_tokens gotcha for thinking models) and why consistency beat average latency. The cron ran this morning. Still $5/month.
Google I/O Challenge → dev.to/dannwaneri/mcp-just-landed-...
Google dropped MCP support in AI Edge Gallery right when I was already running MCP servers on my desktop. Got a Pixel, hit "no eligible devices" on first try, found a second device, and went through the actual setup — model list, skills architecture, MCP connection flow. Wrote what's really there vs. what the announcement implied.
Hermes Agent Challenge → dev.to/dannwaneri/built-a-100k-doc...
For the past six months I've been building what Hermes does by hand — hybrid BM25 + vector search, cross-encoder reranking, MCP server, 100k+ documents. I wanted to know if the tool could actually read my codebase. It described RRF with k=60, all six routing modes, and the Durable Objects backing the MCP server — in 47 seconds. It got every major component right. That was not the result I expected.
Happy Friday 🎉
The emphasis on local, smaller-parameter execution is exactly right. We spend so much time chasing absolute peak benchmarks on massive cloud APIs when, for 90% of local dev loops, a lightweight model that doesn't blow past your daily token budget is the superior developer experience.
It's gunna get even lighter soon. I'm releasing an open-source MCP for web-browsing that uses the AOM instead of DOM that cuts token drain by 80% and gives an 8x speedup vs traditional. I'll drop a post when it's public on git
And yes, I've tested it on smaller agents too, I have an experimental swarm running in podman, using Gemma 4 e2b, works quite well, you can swarm practically anything, booking cheapest flight, finding cheapest gpu prices, etc.
Hi everyone!
My win this week is that I launched a website for my research:
iroha1203.dev/
I’m using it to collect and share notes on the topics I’m currently researching, especially AAT and SFT. In short, my research is about making software evolution computable.
The site is built with plain HTML/CSS and deployed via Cloudflare Pages.
Feel free to check it out!
Are you planning a dark theme? It's easier on the eyes and saves electricity!
Thanks for the comment! I’ve added dark mode support to the site.
Submitted my write up in the last minute for the Gemma 4 challenge.
dev.to/n92/gemma-4-what-i-learned-...
It was a productive week.
Currently reading "Thinking in bets" book
Published article on Frontend HLD: understanding different types of applications
Also made updates to one my projects and finished another project of mine.
I have also decided start posting more often on my blog as well as dev.to
This week I published my submission for the Gemma 4 Challenge 🎉
Built NeuralPocket — a private multimodal AI assistant that runs fully on-device.
No cloud, no servers, no subscription.
✅ Android app + Web version (PWA)
✅ Text, photo and voice input
✅ Works in airplane mode
I've joined several Google hackathons before and always got just a participation
badge 😄 This time I actually want to place!
👉 dev.to/prema_ananda/neuralpocket-p...
MCP and Skill for my agent system seem to be working, and most of the tests are passing. I just need to fix some parts of Skill.md and test it again. I'm planning to share more about it on #githubchallenge
got my project selected in
What was your project. I was planning to participate in it but drop the idea. My institute is not supportive.
It is a deepfake detector. The final round is yet to occur on 30th may
Great. Hope you win.
Thanks
This is somewhat new — a free tool that shows EOL Risk Scores for 8,000+ software versions. Turns out a lot of people are running Debian 12 with 18 days left on the clock and don't know it. Small win, but it feels useful.
endoflife.ai
Hello everyone 👋
This week I finished my submissions for the Hermes Agent Challenge and the Google I/O Writing Challenge.
I enjoyed working on these challenges. Seeing all the different perspectives and discussions after publishing the articles was one of the best parts. It gave me a lot to think about, and overall it has been a really good week.
Once again, thank you to the DEV team for organizing these challenges 😄
github.com/hsr88/mouzi - this is my greates win overall (so far)! Mouzi is causing quite a stir in certain Reddit threads. I already have over 150 stars on GitHub, which is a success for me! 🐭
i get my first SaaS subscription, and successfully earn funding on my company at codeverta
Finally started posting publically about my application and trying to draw users towards it. Been in polish mode for almost 1.5 months and finally said "screw it lets go"
After reading other comments i found myself as very new to the coding like everyone in here is like giant in coding and i am still kid. lol. but i build my own app in android studio after working hard for 2 months and got approval of PlayStore after 14 days of user testing. although app is not upto my expectation and is very basic , it feels me like i achieved something and gives me feel like coder.
/sagar
My win this week was shipping faster without overengineering everything.
Spent more time talking to users, polishing real workflows, and less time rebuilding architecture that nobody would notice yet.
Also finally fixed one backend bug that had been randomly appearing for weeks. That felt better than any productivity metric honestly.
Started work fleshing out a language William James Sidis started when he was 7, Vendergood.
When a language model says "I think X is true," it could mean half a dozen things. Settled knowledge it doesn't want to overclaim. Working belief held under live uncertainty. An inference drawn from what's absent in the evidence. Content recalled from training. A deliberate projection it knows is fiction. A guess filling a gap. In English — and therefore in the training data English produces — all of these collapse into the same hedge. So we built a constructed language called Vendergood — a reconstruction-and-extension of W.J. Sidis's 1898 prodigy-child conlang, with new vocabulary aimed at the distinctions modern AI keeps eliding — as a tool for producing training data that forces the distinctions to the surface. The point isn't for anyone to speak it. The point is to give annotation work and synthetic-data generation a target vocabulary where each kind of "I think" has its own form, so a model trained against it can hold the distinctions structurally rather than fold them into a single hedge.
The load-bearing additions are four epistemic verbs — sciar (certain knowledge), cogitar (working belief), videbar (uncertain perception), imaginar (known-false projection) — and a source-channel system that marks the origin of any claim with one particle: memory, inference, report, dream, deliberate modeling, enchanted frame. Here's a line from our Vendergood translation of Andersen's Snow Queen, where roses tell Gerda that her missing friend Kai is alive:
Incant: nos in terro fuimus — et il non inter ili erat. Ergo il vivit.
In enchantment: we were in the earth — and he was not among them. Therefore he lives. One clause carries three things English can't compress: the source is flagged (Incant: — the roses speak from inside an enchanted garden, where flowers can travel underground and return with intelligence); the inference is drawn from absence (Kai wasn't among the dead, therefore he's alive); the conclusion is direct (vivit — indicative present, no hedge). An LLM trained on text annotated like this would have grammar for the difference between I read it in training and I dreamed it up to fill the gap — because the language itself makes the difference unavoidable.
The repo is public domain (CC0) at libro-vendergood. It currently holds 22 word entries, two grammar volumes, Andersen's Little Match Girl translated in full plus four stories of The Snow Queen, and an Emdros queryable corpus database of every annotated sentence — structured semantic queries via MQL. The load-bearing parts can be learned in an afternoon. The project is CC0 because the point is for the language to find use: train models on it, annotate against it, fork it, ignore the parts that don't fit, build on the parts that do. If you've ever wanted an AI system that can tell you what kind of "I think" it just said — and which parts of its answer it's actually hedging — this is one way to give it the vocabulary.
Shipped two protocol features for NOVAInetwork
(AI-native L1 blockchain in Rust): multi-party
payment splitting (one payment atomically credits
multiple providers) and agent upgrade (swap an AI
agent's model version without losing identity,
reputation, or open payment channels). 115 new
tests this week, 1,768 total. Solo founder,
building since December 2025.
Made some more changes to my first ever deployed web app!
And finished up the basic structure for my Gemma4 Challenge submission! Now i just have to get the demo video recorded and my post all wrote up to submit! Depending on how it goes on here, I might be confident enough to pitch it to my current employer to be used at my job lol Have a great weekend everyone!
Finally understood a concept that confused me for days 😅 Small progress is still progress!
A small one but it made my week. I left a quibble on a multi-tenancy post about a missing RLS performance section, and the author actually went back and added a whole new
section to address it, then pinged me to recheck. Small moment, but exactly the kind of platform-level engagement that makes commenting on dev.to feel worth doing well rather
than scrolling past.
Also shipped a "two-layer" governance doc for my open-source LMS (open code, curated hosted brand) so future contributors don't have to guess what contributing actually gets
them. Writing the rule down was the most useful part.
Finally started drafting up ideas to get users for my Ez Garden Visualizer app. It's scary - I much prefer to just keep working on the app! But talking about it earlier will help getting feedbacks much more quickly!
I will now
The mindset change itself I think is a win for me ! 😊
What an inspiring thread! 🎉 This week, my win was pushing through a challenging project and finally seeing it come together. Reading about everyone's achievements here — from building first apps to finishing a teaching semester — is a great reminder that consistency and courage pay off. Keep it up, everyone! 💪
Done with audience pages for DevIntern, trying to speak directly to people instead of making one generic pitch is hard
Travelling to Madrid to attend the MadVue conference and gathered speakers for September's PragVue conference of mine.
Getting my Idea ready for alpha launch on july 4th ! Dev.to article coming up !
I coined a term I hope will stick Cascading Markdown Sheets. I built MagpieCSS to handle not only making my projects look good, but to also work hand and hand with the css to convert all pages to markdown for easier ingestion by LLMs! It's been a blast.
What an absolute banger of a week! 🔥 Reading everyone's wins is giving me so much motivation. I just crossed the finish line and submitted my projects for both the Gemma 4 Challenge and the Hermes Agent Challenge! It was a wild ride juggling both, but seeing what this community is building makes it all worth it. Huge congrats to everyone shipping code this week! 🙋♀️😊✨
I am proud of having finished and published my open-source community-driven website, VocaWeb. A website for Vocaloids and their music. More features to come(after finishing the music search system) , I would love some feedback
here is the GitHub repo: repo
And the live server: github pages
I hope u enjoy it
my article was published on freecodecamp freecodecamp.org/news/build-autono...,
my project OpenOSINT/OpenOSINT got sponsored by ip2location.
Reached 300 followers on linkedin.
Made a bunch of improvements to FiveClues this week 🧩 : fiveclues.live
It’s a small daily game where you guess an AI tool from clues. Kind of like Wordle for AI products.
Honestly, the biggest win was finally shipping an idea that sat in my notes for almost an year.
Shipped VibeSec — an open source security scanner for AI-generated code from Cursor, Bolt, and Claude Code.
Built it in one day: 10 security rules, 39 passing tests, published to PyPI.
Ran it on my own projects and found 4 vulnerabilities in my own code.
pip install vibesec
I am proud to have launched my first OSS package, SirenSpec. Have a full write-up coming Wednesday!
Repo: github.com/sirenspec/sirenspec
Docs: docs.sirenspec.dev
Mine was new article here.
I finally got this post out about the similarities between accessibility and devrel! dev.to/andyhaskell/accessibility-t...
hi dear can you spare few minutes here"dev.to/halakabir234hub/looking-for..."
guys check my profile and please do react i'm very close to winn
got out of a 2-week slump to finally study for exams. hand(!!!) coded a lil sudoku solver after reading that it's a graph colouring problem
Well. Despite some turbulence in personal life — shipped two MVP-level learning projects and kept moving. 🙈 💪
Love this. My win this week was learning through the process, even from the confusing parts. Small progress still counts, and that dog at the laptop is basically my Friday mood 😂
I solved "Maximum Water Contained" LeetCode array problem, well the question itself was kind of tricky at first but once i found out formula of area it was easy.
I solved "Most Water Contained" array problem on LeetCode. well the question itself was tricky at first but once I figured out its just formula of area it was easy.
Got my next reading list sorted out. I have read 1 book a month for a year it was challenging for a long while but now its becoming easier.
Finally pushed my InternalVoice project to Github. Give users true ability to interact with their OS and Hardware, using STT and TTS.
Finally got my PR merged after debugging for hours. Those small victories hit different!
🤔 I am as new entry in cloud hybrid platform and make next step for dual boot display as end - user.
Hehehe 🦅🦅🦅🦅🫡
Thank you
Finally understood a concept I’d been struggling with for weeks while learning programming. Small win, but felt great 😄
I'm proud that I just released my new app that could potentially help lots of students in my country!
cool!
I went live with RepoSignal.io. We identify risky PR's based on your repo's history. It is definitely not just a run of the mill scanner.