Inspiration
Hamouda joy is inspired by my grandfather who was a blind person working in a factory. He has an exceptional ability to calculate and manage money and give answers to complex analysis. He was my source of inspiration since I passed all my youth with him. It's for him that I make this small project.
What it does
Hamouda joy helps blind people or with low vision to turn local CSV data into clear answers, spoken insights, accessible chart explanations, and personalized reports. Data stays on the user’s device. Users can explore their data through supported questions, compare results, hear answers, examine exact calculations, and generate an accessible report based on their session. All processing happens locally, so the data stays on the user’s device.
How we built it
I built Hamouda Joy with dependency-free HTML, CSS, and JavaScript. It uses the FileReader API for local CSV files, inline SVG for visual charts, semantic HTML for accessible chart explanations, and the browser’s Web Speech API for optional spoken output. I used Codex CLI with GPT-5.6 to implement, refactor, test, audit, and document the application. I also used ChatGPT and Codex Desktop with GPT-5.6 to develop the accessibility strategy, deterministic analysis workflow, experiment, and demo narrative. I reviewed each milestone and explicitly approved important actions such as commits and pushes.
Challenges we ran into
The largest limitation is that the prototype has not yet been tested by blind or low-vision professionals. I performed initial creator testing using keyboard navigation, Microsoft Edge, Windows Narrator, 200% zoom, spoken output, and privacy inspection. This is not accessibility certification or a replacement for testing with the intended users. I also compared Hamouda Joy with a manual Excel workflow using two small exploratory tasks. In this single creator experiment, Hamouda Joy was approximately 13.6 times as fast for one task and 19.9 times as fast for the other. These early results are encouraging, but they cannot be generalized because the experiment involved one tester, two tasks, approximate Excel timings, and no independent observer.
Accomplishments that we're proud of
I have got a working project, good implementation of agents in a short time. A solution that can improve and help blind people in different sectors. After all we got a project that analyses CSV, generate charts, answer questions and generate report based on user exploration and questions. It also doesn't give random chart or answers, and it keeps context and even if you ask a wrong question, it will guide you to ask the right ones.
What we learned
I learned how to use codex, ChatGPT, optimize usage and create a full application from scratch
What's next for Hamouda Joy
It will be test with blind data professionals and see its impact on them and how it can help them in their daily work. It will support factory operations budgets, and additional chart types. Future releases can have optional voice questions stronger data validation ...
Log in or sign up for Devpost to join the conversation.