The latest articles on DEV Community by Sonik (@solt_10). https://dev.to/solt_10 https://media2.dev.to/dynamic/image/width=90,height=90,fit=cover,gravity=auto,format=auto/https:%2F%2Fdev-to-uploads.s3.us-east-2.amazonaws.com%2Fuploads%2Fuser%2Fprofile_image%2F4180908%2Fb459e828-8425-409f-8bfd-8b485f6fbf54.png https://dev.to/solt_10 en Sonik Tue, 13 Oct 2026 14:01:15 +0000 https://dev.to/solt_10/accesssprint-a-local-accessibility-review-side-hustle-with-gemma-4-1ecp https://dev.to/solt_10/accesssprint-a-local-accessibility-review-side-hustle-with-gemma-4-1ecp <p><em>This is a submission for the <a href="https://dev.to/challenges/hacktoberfest-week2-2026-10-12">Hacktoberfest Open-Source AI Challenge Week 2: Side Hustle</a></em></p> <h2> What I Built </h2> <p>AccessSprint turns a small, clearly scoped website scan into a human-reviewed accessibility action plan. It is for freelance developers who want to offer independent businesses a practical review: prioritized findings, notes, hands-on checks, and a report they can use to plan fixes.</p> <p>The service is intentionally small: review up to 10 public pages, walk through the evidence with the client, and offer a follow-up scan after fixes. AccessSprint keeps automated scanner results separate from human observations. A model can explain a finding, but it cannot confirm, dismiss, or fix it for you.</p> <h2> Demo </h2> <p>A video walkthrough is coming soon. The app currently runs locally on Windows; the README has setup steps and troubleshooting:</p> <ul> <li><a href="https://github.com/soniksharma/ASprints#readme" rel="noopener noreferrer">AccessSprint setup guide</a></li> <li>Start the app with the included Windows launcher, then open its local 127.0.0.1 address.</li> </ul> <h2> Code </h2> <p><a href="https://github.com/soniksharma/ASprints" rel="noopener noreferrer">Source code and README on GitHub</a></p> <h2> How I Built It </h2> <p>AccessSprint combines a small Python and SQLite backend with a JavaScript interface. A local Playwright/Chromium worker runs axe-core against only the public URLs the auditor explicitly adds. The app saves per-page results, review notes, manual checks, scan comparisons, printable reports, and versioned JSON backups.</p> <p>For explanations, it calls Ollama on loopback with my local Gemma 4 gemma4:e4b model. The prompt contains a selected rule and bounded, sanitized evidence—not the client name, project notes, or page HTML. The explanation returns a plain-language impact, repair suggestion, and verification step. The human auditor remains responsible for reviewing the result.</p> <h2> Why Does Open Innovation Matter? </h2> <p>A freelance service needs predictable costs. Running Gemma locally means there is no hosted model charge every time I explain a finding, which makes small, fixed-price reviews easier to offer. Local inference also keeps client context on the auditor’s device and makes the tool usable without sending audit evidence to a third-party AI API.</p> <p>Open tools make the workflow inspectable and replaceable: axe-core provides the automated checks, and Ollama lets the auditor choose a local model. The trade-off is that users need a compatible computer and must install the model themselves.</p> <h2> First-user status </h2> <p>The prototype is built, but I have not yet validated it with a paying client. My next step is to put it in front of an independent developer or small business owner and learn whether the report is clear enough to support a real paid review.</p> <h2> Prize Category </h2> <ul> <li>Best Use of Gemma</li> </ul> <h2> Important limitation </h2> <p>AccessSprint is an audit aid, not a WCAG certification or legal opinion. Automated checks cover only part of accessibility; every finding and key user flow still needs human review.</p> devchallenge hf26challenge gemma opensource