SafeCheck

Inspiration

Scammers don't target one age group; they target everyone in different ways. Older adults get fake bank alerts, "grandchild in trouble" calls, and tech-support scams. Younger people get fake job offers, payment app tricks, marketplace scams, and gaming account takeovers. Yet most scam-checking tools are either too technical for older adults or too slow and cluttered for younger users. We wanted to build one tool anyone could trust, whether they're 16 or 80, that answers two questions: "Is this a scam?" and "What do I do now?"

What it does

SafeCheck lets anyone paste a suspicious message, upload a screenshot, check a link, or scan a QR code, and instantly get a plain-language verdict: "Likely a scam," "Be careful," or "Looks safe." It highlights the exact red flags in the message, explains each one in simple words, and walks the user through what to do next.

Two experience modes: Simple mode guides older adults one step at a time with large controls and read-aloud. Quick mode shows everything on one screen with keyboard shortcuts for faster, tech-comfortable users. Both share one engine, and users can switch anytime without losing their work. Link and QR code inspector: It reveals where a link really goes and flags lookalike domains, hidden shorteners, and disguised characters, without ever opening the link. Phone call coach: For someone on a suspicious call right now, they tap what the caller is asking for and instantly get a verdict and a polite script to hang up safely. It works without AI, so it's instant. "What should I do now?" guide: tailored step-by-step plans, including a path for people who already clicked a link or shared information. Scam help for every generation: from grandparent and bank scams to fake internships, payment app tricks, gaming account takeovers, crypto offers, and sextortion threats, each with calm, tailored next steps. Spot the Scam practice mode: a quiz with realistic messages, scores, streaks, and a shareable result. Family safety plan: a guided setup for a family code word to beat AI voice-cloning scams, with a printable card. Auto-redaction: Card numbers, Social Security numbers, account numbers, and verification codes are hidden before anything is sent for analysis, and users see what was hidden. Ask someone I trust: It creates a plain-language summary to share with a family member. English and Spanish support, including AI results in the chosen language. Accessibility built in: text size control, high-contrast mode, reduced motion, read-aloud, full keyboard navigation, and screen-reader support. How we built it

We built SafeCheck with React and Vite on the frontend and a small Node/Express server that sends messages and screenshots to the Groq API for analysis, keeping the API key securely on the server. The AI returns structured results, including the verdict, exact quoted red flags, and next steps, which the app uses to highlight suspicious phrases directly in the original message. Both modes run on one shared engine and state. QR codes are decoded in the browser with jsQR, link inspection runs locally against a list of commonly impersonated brands, and private information is redacted in the browser before any request is sent. Read-aloud uses the Web Speech API, and sharing uses the Web Share API. A demo mode with pre-written results keeps the app working even without an API key. We used Claude Code as an AI coding assistant to help build quickly, while we focused on the users, flows, and design decisions.

Challenges we ran into Designing for two very different users without compromising either. Adding speed for younger users risked overwhelming older ones. Two modes on one shared engine solved this. Making AI results reliable. Getting consistent, structured answers and matching quoted red flags back to the original message took careful prompting and fallback handling. Writing in truly plain language. Explaining things like lookalike domains at a sixth-grade reading level, in two languages, was harder than expected. Handling sensitive situations with care. Scams like sextortion required calm, non-judgmental guidance that protects users without frightening them. Keeping data safe. We kept the API key off the browser, redact private details before analysis, and never open the links we inspect. Accomplishments that we're proud of One app that works for both a grandparent and a teenager, with each getting an experience designed for them. Highlighting the exact suspicious words, so users learn why something is a scam instead of just being told. A phone call coach that helps people in the moment, when scams are most dangerous. Privacy-first design with automatic redaction. Accessibility treated as a core feature from the start, not an afterthought. What we learned

Good UX starts with understanding the person, not the feature list. Designing for two very different users taught us that the same information can need completely different experiences. Small decisions, like showing one step at a time and using reassuring language instead of alarming warnings, make a huge difference. We also learned to work with AI responsibly: structuring its output, planning for when it fails, and protecting keys and user data.

What's next for SafeCheck User testing with older adults and students to refine both modes. More languages beyond English and Spanish. A browser extension and mobile app to check messages and links where people actually receive them. Partnerships with libraries, senior centers, and schools to bring scam-awareness training to their communities.