SShortSingh.
Back to feed

Publish Gates Blocked 23 Flawed Posts in 10 Days for a Claude Code Blog Pipeline

0
·3 views

A developer running a Claude Code-driven blog pipeline placed a Python gate between the model and publishing, checking actual script output rather than trusting the model's success reports. Over 10 days, the gate ran 248 times and blocked 37 runs, stopping 23 distinct posts from going live. The blocks came from mechanical defects such as missing keyword research, charts not embedded in the post body, and stale relative dates. The developer built the rules after failures that slipped past prior checks, including 132 posts published without images. Results were recorded in append-only JSON ledgers written on every run.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingDEV Community ·

Houdini Pretrained ML Volume Upres Tested on Pyro Smoke, Plus a Small Training Run

A Digital Garden cook log describes using Houdini 22.0.429's pretrained ML Volume Upres nodes to increase the spatial resolution of simulated billowy smoke. The author ran the native billowy-smoke recipe on a licensed Windows workstation with an NVIDIA RTX 5070 Ti, producing valid 2x and 3x density volumes. The post stresses that the upscaling model sharpens a cached field but does not change the upstream simulation's motion or dynamics. A short native training run also exported an ONNX model that could be reapplied through the same workflow. The author reports that the main difficulties involved field selection, environment setup, and the limits of what a brief training run can demonstrate.

0
ProgrammingDEV Community ·

Developer builds free browser-only site of playful Gen Z text and novelty tools

A developer has launched a side project called My Gen Z Tools, a collection of novelty calculators and text generators. Inspired by trends on TikTok and Instagram, the creator wanted to avoid paid APIs, backend servers, and databases, keeping hosting free and fast. The site uses vanilla JavaScript that maps standard letters to Unicode symbols to produce decorative text, and it computes a name-based "Vibe Score" locally in the browser. The developer says the project shows that browsers can handle fun, viral tools without heavy backends, and is now seeking SEO advice, traffic, and feedback on the interface and code.

0
ProgrammingDEV Community ·

Keeping Presentation Slides in Git With Tagged PPTX Exports for Each Release

Teams often struggle with presentation files piling up as confusingly named copies, which makes changes hard to track and past versions hard to recover. The article proposes storing slides as text-based HTML or Markdown so version control tools can show line-by-line edits. It recommends a simple repository layout, branches for audience-specific variants, and code-style reviews for slide changes. Each presented or externally shared version should be tagged, then exported to PDF or PPTX as a release artifact rather than edited directly. Automating previews and PDF exports can further reduce manual work, while large assets should be compressed to keep the repository efficient.

0
ProgrammingDEV Community ·

Open-source AI agent flags a water spike, duplicate charge, gym fee hike, and forgotten subscription

A developer built Napper, an offline-first AI agent that processes locally dropped CSV files of bank, electricity, step, and water data overnight. Using the open-weight TabPFN-3.5 model for forecasting and several detectors for anomalies, it produces a short morning brief delivered by text and audio. In a test with synthetic data, it identified a water usage spike of 425 liters on October 19 against an expected 52.8 liters, and a 262% electricity surge in September. It also caught a gym membership fee rising from $24.99 to $29.99, a duplicate $77.86 power charge on the same day, and a $9.99 monthly streaming subscription unused since March. The project was submitted to the Hacktoberfest Open-Source AI Challenge and uses synthetic datasets that anyone can reproduce.