THE PROBLEM Many people in Nigeria (housemaids, plumbers, market traders, older people) are asked to sign agreements they cannot fully read. Signing something they don't understand can lock them into a deal that takes their time and energy.
WHAT IT DOES IML lets anyone snap a photo of an agreement, choose Hausa, Igbo, Yoruba or Pidgin, and get a clause-by-clause explanation with the original text beside it. A voice player reads the explanation aloud (pause, back, next, replay). Tick boxes are explained with short, basic suggestions. If a photo is blurry or the document looks cut off, IML says so instead of guessing. Explanations are saved only on the user's phone for 30 days, and the photo is never kept.
WHO IT HELPS People with little formal education who do odd jobs or trade, older people who read slowly, and the relatives who help them.
HOW I BUILT IT A mobile web app (HTML, CSS, JavaScript) with a small Python Flask server. The server sends the photo to a vision-capable AI model through OpenRouter, using strict accuracy rules: explain every clause in order, keep the original text, never guess, flag unreadable or incomplete pages. Voice uses the browser's speech synthesis. Hosted free on Render, code on GitHub.
CHALLENGES Phone voices for Hausa, Igbo and Yoruba are often missing or unclear, while Pidgin works well. The free hosting plan makes reading a document slow (up to about 2 minutes). The explanations in each language still need review by native speakers.
WHAT I LEARNED I learned how the parts of an AI app connect: a photo goes to a server, the server sends it to an AI model, and the explanation comes back to the phone. I learned that an API key and a model are two separate settings, that a free hosting plan can be slow and needs a longer timeout (I fixed a failed request by raising it to 180 seconds), and how to take a project from my laptop to GitHub to a live link. The biggest lesson was accuracy: the AI has to be told clearly never to guess, and to say when a page is cut off or unreadable.
WHAT'S NEXT PDF upload, downloadable voice notes and PDF copies of explanations, better voices for Hausa, Igbo and Yoruba, and native-speaker review.
MY CONTRIBUTION AND AI TOOLS I designed the product (users, languages, accuracy rules, voice player, privacy choices), tested it on real agreements, and set up, deployed and submitted it. I used Claude (Anthropic) to help write the code from my specification and to guide me through setup. The explanations are generated by an AI model through OpenRouter.
Built With
- css
- flask
- html
- javascript
- openrouter
- python
- render
- web-speech-api
