DEV Community

Cover image for Is This Real? A Forward Checker I Built for My Parents
Anish Kumar
Anish Kumar

Posted on AI-assisted

Is This Real? A Forward Checker I Built for My Parents

#devchallenge #weekendchallenge #hf26challenge

Hacktoberfest Weekend Challenge: Build for a Friend Submission 🤝

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend

What I Built

Every family group chat has that one message. A government scheme that pays out if you click a link. Hot water with lemon that cures a serious illness. A text saying the bank account is locked and you must share a code right now.

My parents get these every week. When they ask me, it takes me two minutes to spot the scam. But I'm not sitting next to them every time a message lands, so I built them something that is.

Is This Real? is one page. You paste a forward or a text, press one big button, and get:

  • A verdict: SCAM, FALSE, MISLEADING, UNVERIFIED, or LIKELY TRUE
  • A one-line explanation in plain English
  • Red flags it found, like urgency, a strange link, or a request for a one-time code
  • What to do next, for example "Don't click. Call your bank on the number printed on your card."
  • Whether it's safe to forward, because stopping the chain matters as much as not falling for it
  • A Read aloud button, because small text on a phone is hard for older eyes

The text is large and the layout is plain on purpose. It's built for someone who is 65, not for me. And scams look the same everywhere, so it works for parents in any country: fake parcel fees, "Hi Mum, this is my new number", prize wins, fake account alerts.

Demo

Try it here: https://sach-ya-jhooth.vercel.app

Tap one of the Try examples on the page to test it in one click, or paste your own message.

Code

GitHub logo Sarcastic-Soul / sach-ya-jhooth

Sach Ya Jhooth? (True or False?)

Paste a WhatsApp forward or SMS and get a plain verdict in English and Hindi: scam, false, misleading, unverified, or likely true. Built for my parents, who get these messages every day.

Live: https://sach-ya-jhooth.vercel.app

How it works

  • index.html: one page, no framework.
  • api/check.js: a Vercel serverless function that sends the message to gpt-oss-120b, an open-weight model (Apache 2.0), served by Groq. The model returns JSON with a verdict, a summary in English and Hindi, red flags, what to do, and whether it's safe to forward.
  • Messages are not stored.

The model can't browse the web, so for news it says "unverified" and points to PIB Fact Check.

Run it yourself

npm i -g vercel
GROQ_API_KEY=your_key vercel dev
Enter fullscreen mode Exit fullscreen mode

Because the model is open-weight, you can swap Groq for any OpenAI-compatible server (Ollama, vLLM, llama.cpp) by changing the URL in api/check.js.

License

…
View on GitHub

How I Built It

The whole project is two files.

  • index.html is a plain HTML page with a little vanilla JavaScript. No framework, no build step.
  • api/check.js is a Vercel serverless function. It sends the message to openai/gpt-oss-120b, an open-weight model released under Apache 2.0, served through Groq's free tier.
  • Read aloud uses the browser's built-in speech (Web Speech API), so it's free and needs no extra service.

The system prompt tells the model who it's helping (an older parent) and what to look for: fake bank alerts, prize and lottery wins, payment fraud, one-time code requests, parcel delivery fees, "new number" scams, fake government schemes, health myths, edited quotes, and old news shared as new.

I ask the model for JSON only, using the JSON response mode, with fields for the verdict, a summary, red flags, the next step and a safe-to-forward flag. The front end just fills in a card. Temperature is low (0.2) so the answers stay steady.

I was also honest with the model about its limits. It can't browse the web, so if a claim depends on recent news it should say UNVERIFIED and point to a fact-check site. The page footer links to Snopes and Reuters Fact Check for the same reason. Messages aren't stored anywhere.

What's Next

  • A WhatsApp bot, so parents can forward a message straight to it instead of opening a website
  • Their own language: one prompt change gives Spanish, Hindi or Portuguese answers
  • Fully local: run gpt-oss on a home machine with Ollama so messages never leave the house

Why Does Open Innovation Matter?

The messages my parents get are private. Some contain their bank name, their phone number, or a family member's details. With an open-weight model, nothing forces those messages to go to one closed company. The same app could run gpt-oss-120b on any provider, or on a machine at home, and the code wouldn't change.

It's also free. Groq's free tier serves the open model, Vercel hosts the page, and the code is open source. A tool for parents shouldn't need a subscription.

The Apache 2.0 license matters too. Anyone can fork this and change one line in the prompt to answer in their family's language, without asking anyone's permission.

With a closed API, I'd be renting my parents' safety from someone else's pricing page. With open weights, it stays ours.

My Agent Session

I built this with Claude Code in a single session under a 20-minute deadline: idea, code, deploy to Vercel and this post. The open-weight model made that possible, since there was no waitlist, no approval and no card to add.

Top comments (0)