Self-auditing AI QA agent logs 79 findings against its own code in nightly reviews

·5 views

A developer built an AI-powered QA agent called Verdict that audits its own releases every night rather than reviewing external codebases from scratch each time. Unlike typical AI QA tools, Verdict maintains a persistent state file, a failure taxonomy, and a signed run history, and is strictly prevented from editing the code it evaluates. The system classifies every failing test into categories such as real defect, brittle, or flaky, and assigns each finding a stable ID and verdict from a fixed set of outcomes. Over its operational period, Verdict has filed 79 findings against its own harness, including repeated defects in fixes for earlier issues, prompting a process change requiring fix verification along untested axes. When run against the open-source library itsdangerous on a fresh clone, it completed an audit in 18 minutes, flagging nine findings including a major gap in timing-attack defences and an unpinned boundary condition in token age validation.

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)

Engineer shares 18 lessons from building 12 AI voice models using diffusion TTS

A software engineer documented roughly one month of work designing and deploying 12 role-specific AI voice models — covering narrator, counselor, sales, presenter, operator, and MC roles for both men and women. The project used diffusion TTS for voice design, where each voice is defined by a text caption and a random seed, making results fully reproducible. A key finding was that diffusion TTS ran 2.5 times slower than real time on the same GPU, making it unsuitable for live conversation and prompting a two-stage approach using lighter pre-trained models for actual speech synthesis. The engineer identified corpus quality gates as a central problem, noting that six of the 18 documented failures involved gates that existed but did not work as intended — for instance, strict gates inadvertently filtered out expressive takes, leaving only monotonous audio. Additional pitfalls included undetectable long-vowel defects caused by text normalization, a hallucination guard that only failed during actual hallucinations, and symbol-stripping rules that silently dropped valid Japanese characters from input text.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Builds Interactive Excel Dashboard to Analyse Jumia E-commerce Data

A developer has created an interactive Excel dashboard using transactional data from Jumia, one of Africa's largest e-commerce platforms. The project involved cleaning raw data by removing duplicates, fixing inconsistencies, and handling missing values before analysis. Excel functions such as IF, AND, AVERAGE, COUNTA, MIN, and MAX were used to categorise prices, discounts, ratings, and calculate key metrics. Pivot tables, charts, and slicers were incorporated to enable dynamic filtering and visual data exploration. The analysis concluded that consistent product quality and customer value matter more to sales performance than high discounts or large review volumes alone.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Excel Dashboard Built to Analyze Jumia Product Pricing, Discounts and Ratings

A data analyst used Microsoft Excel and Power Query to clean and visualize a 115-row Jumia e-commerce dataset covering product prices, discounts, ratings, and reviews. The raw data contained multiple quality issues, including 58 blank review and rating fields, negative review counts, price ranges instead of single values, and ratings stored as text strings. Cleaning steps included converting negative review figures to absolute values, extracting midpoint prices from ranges, stripping currency symbols, and standardizing rating formats. The analyst deliberately left missing review fields blank rather than replacing them with zeros, to distinguish unavailable data from confirmed zero-review products. The finished interactive dashboard — built with PivotTables, PivotCharts, and slicers — was designed to answer business questions such as whether higher discounts drive more customer engagement and whether price correlates with product ratings.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

AuthGeek offers offline TOTP authentication with Argon2-encrypted local vault

A developer has released AuthGeek, a free open-source desktop authenticator for Windows that stores TOTP and HOTP codes entirely on the local machine with no cloud sync or account required. Secrets are secured in a local vault using Argon2id key derivation, chosen for its memory-hard properties over alternatives like PBKDF2. The app supports adding accounts via QR code scan or manual secret entry, and includes encrypted backup and restore functionality. Built with .NET 8 and Avalonia UI, the tool is available on GitHub, though the installer is not yet code-signed and may trigger a SmartScreen warning on first run. The developer openly acknowledges that storing codes on the same machine used for login is a weaker security model than a separate physical device, and advises users with stricter threat models to continue using a phone-based authenticator.

0 comments Read more at DEV Community