ELI10 — Explain Like I'm 10

Inspiration

Almost everyone gets a document they don't fully understand: a hospital discharge leaflet, a tenancy clause, a school trip form. Often it's someone helping a parent or relative who ends up reading it. Pasting it into a chatbot gives you a wall of text, and you can't tell which parts came from the document and which parts the AI made up. I wanted a simple explanation you can actually check.

What it does

You paste a confusing document, or try one of three samples, and press Explain it. ELI10 returns a short card with four parts:

• In short: what the document is about, in 2–3 plain sentences
• What you need to do: the actions, with deadline badges such as "within 2 months"
• Watch out for: risks and easily missed details
• Questions to ask: ready to take to the doctor, landlord or school, and copied with one click

Every point comes with the exact quote it was based on. Click a point and that sentence lights up in the original text. If the AI's quote can't be found in the original, the app says so instead of showing a false highlight.

Nothing is stored, and the app explains documents without giving legal or medical advice.

How we built it

• I followed the Devpost Learn skill pack step by step: Start → Scope → PRD → Spec → Build → Ship. Each stage produced a planning document before any code was written.
• The server is plain Node.js with no npm dependencies, and the front end is plain HTML, CSS and JavaScript.
• Google Gemini returns a fixed JSON format, and every point has to include a source_quote taken from the original text.
• A matcher checks each quote against the original. That check is what makes the highlighting trustworthy.

Challenges we ran into

• Gemini sometimes returned "high demand" errors, and once it hung for several minutes. I added a 30-second timeout so the app fails cleanly instead of spinning forever.
• The free quota on the main model ran out during testing, so I added fallbacks to lighter Gemini models.
• The AI's quotes had to match the original word for word, and they didn't always. I wrote a check script that runs the samples and measures how many quotes are actually found in the source.

Accomplishments that we're proud of

Clicking a point and seeing the exact original sentence light up. It turns "trust me" into "see for yourself." I'm also pleased that the whole thing runs with no dependencies and stays honest when a quote doesn't match.

What we learned

Planning before building felt slow at first, but having a clear scope and spec made the build much faster and kept me from adding things that didn't matter. I was also surprised by how often an AI "quote" isn't quite word for word. You have to check it rather than assume it.

What's next for ELI10 — Explain Like I'm 10

• Upload a photo or PDF of a letter instead of pasting the text
• Translate the explanation into other languages, which would help families where English isn't the first language
• A reading-level slider (age 7 / 10 / adult)
• Read-aloud for people who find long text hard going

Built With

• css3
• gemini-api
• google-gemini
• html5
• javascript
• node.js