Developer builds REST API to audit GitHub repos and score issue complexity with AI

·1 views

A developer has launched RepoAudit, a REST API that audits any public GitHub repository and returns a data quality report alongside AI-powered complexity scores for individual issues. The tool addresses a common pain point for open source maintainers — disorganised issue trackers with missing labels, stale tickets, and no effort estimates for contributors. Built with TypeScript and Express, the API separates deterministic quality checks from AI scoring via Groq's Llama 3.3 70B model, ensuring partial results are still returned if the AI layer fails. Key lessons during development included capping AI scoring at 20 issues per full-repo audit to reduce latency, and configuring proxy trust settings on Render to fix IP-based rate limiting. The API is live with a free tier of 10 audits per day, alongside paid plans at $9 and $29 per month.

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)

Why Technical SEO Audits Should Come Before Creating New Content

A technical SEO practitioner argues that publishing more content is often premature if a site's existing pages cannot be properly crawled or indexed by search engines. The recommended audit process covers HTTP response codes, robots directives, canonical tags, and JavaScript-rendered markup to ensure all signals are consistent. Sitemaps should list only canonical, indexable URLs returning a 200 status, and lastmod timestamps should only update when content genuinely changes. Internal linking is also examined for click depth, broken links, orphaned pages, and uninformative anchor text. Structured data is validated not just for technical correctness but for truthfulness, ensuring markup properties match what is actually visible on the page.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer runs 1,699-record academic site on JSON files, uncovers silent data bugs

A developer has been running ryzenstudy.com, an exam paper repository with 1,699 records, using flat JSON files instead of a traditional database for several months. The Next.js 15 app stores all data in a single file under one megabyte, with every query handled via JavaScript array filters and every write via a synchronous file operation. After months of smooth operation, the developer discovered 95 uploaded papers were effectively unreachable due to duplicate URL slugs generated from subject name and year alone. The root cause was that the same subject, such as Deep Learning, is issued under different course codes for different university branches, producing identical slugs for distinct exam papers. Unlike a relational database, which would have enforced a unique constraint and flagged the conflict immediately, the JSON approach silently allowed duplicate slugs to accumulate, requiring a custom collision-aware slug generation fix.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How to Design Human-AI Oversight That Avoids Approval Fatigue in Enterprise Systems

A new guide on enterprise agentic AI warns that routing every AI action to a human for approval creates 'rubber stamping' rather than genuine oversight, as reviewers quickly succumb to fatigue and approve requests without scrutiny. The recommended approach uses confidence and impact thresholds to triage outputs, so only genuinely uncertain or high-stakes actions reach human reviewers. Presenting reviewers with concise diffs instead of full documents is suggested to cut review time from minutes to seconds, making sustained attention more realistic. An LLM-based auditor can monitor random samples of automated outputs daily, automatically raising human-review rates if quality scores decline. Human corrections and approvals are also treated as labeled training data, feeding back into evaluation pipelines to continuously improve the system.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Reliable Enterprise AI Comes Down to Boring Engineering, Not Model Magic

A technical guide on deploying agentic AI in production argues that reliability hinges on foundational engineering practices rather than model sophistication. The piece recommends full observability through tracing tools, structured testing pipelines, and strict JSON schema validation to keep AI workflows predictable. It emphasizes designing for failure at every layer, including retries with backoff, step limits, idempotent tool calls, and checkpoint-based workflow resumption. Cost controls must be enforced at granular levels — per user, tenant, and account — to prevent runaway cloud spending from buggy retry loops. The author concludes that the harness surrounding the model, covering schemas, limits, secrets, and rollback plans, is the real product, and teams should identify and fix their most critical gap immediately.

0 comments Read more at DEV Community