RecallRadar — Know Before It's Too Late

Millions of products get recalled worldwide. Nobody checks. RecallRadar watches for you, adds a receipt, we alert you instantly if anything you own is dangerous. Free. Automatic. Life-saving.

Comment

RecallRadar

What Inspired Us

One day, a family member bought a baby formula from a local store. A few weeks later, we accidentally found out online that the same brand had been recalled due to contamination.

That moment scared us.

We realized the information was out there. The government had published the recall. But nobody told us. There was no alert, no notification, no warning. We only found out by accident.

That is when we asked ourselves:

"Why does nobody build a system that just... watches for you?"

That question became RecallRadar.

How We Built It

We wanted to keep everything simple and free, no paid APIs, no expensive services, just smart use of what is already available.

Here is how the pieces fit together:

Step 1: Getting the recall data We discovered that governments publish recall data for free through official APIs. The FDA covers food and medicine. NHTSA covers vehicles. CPSC covers consumer products like electronics and toys. We connected to all three.

Step 2: Reading receipts with AI Instead of asking users to type product names manually, we used Google Gemini Flash (free tier) to read receipt photos. The user snaps a photo, and the AI extracts every product automatically in seconds.

Step 3: Matching products to recalls This was the hardest part. A receipt might say "Samsung 65W Charger." But a recall notice might say "Samsung fast charging adapter." They mean the same thing, but a simple text search would miss it.

We solved this using Sentence Transformers, an AI model that understands meaning, not just exact words. It converts both texts into numbers and measures how similar they are:

$$ \text{similarity} = \frac{A \cdot B}{|A| \times |B|} $$

If the similarity score is above 0.6, we flag it as a match.

Step 4: Alerting the user Once a match is found, we send an email via Gmail SMTP with:

  • The product name
  • What is the danger
  • The severity level (High / Medium / Low)
  • Exactly what to do

Step 5: Running automatically A background scheduler checks every user's products every 24 hours, without anyone having to hit a button.

What We Learned

  • Government APIs are powerful and free. We never knew how much safety data is publicly available and just waiting to be used.

  • AI does not have to be expensive. Our entire AI layer runs on free tiers and open-source models.

  • The UX matters more than the tech. The receipt scanning feature was the most important decision we made. If users had to type products manually, nobody would use it.

  • Semantic search beats keyword search. Cosine similarity with embeddings is far more effective than simple string matching for real-world product names.

Challenges We Faced

Challenge 1: Government APIs are inconsistent The three APIs (FDA, NHTSA, CPSC) all have different formats, different field names, and different response structures. We had to build a normalizer to convert all of them into one standard format before matching.

Challenge 2: Product names vary wildly A single product can be described in dozens of ways on a receipt. "Tylenol Extra Strength 500mg Caplets 100ct" vs "TYLENOL 500MG" are the same thing. Getting the matching threshold right took a lot of trial and error.

Challenge 3: Making it feel simple The backend is complex, with three APIs, AI embeddings, a scheduler, email alerts, and a database. But the user should feel none of that. Hiding all the complexity behind a single "snap your receipt" button was a real design challenge.

Challenge 4: Keeping it completely free Every time we found a solution, we asked: "Does this cost money?" If yes, we found an alternative. This constraint actually made us more creative and the product more accessible to everyone.

Why It Matters

Every single day, people get hurt by products they did not know were recalled. Parents feed their children recalled baby formula. Drivers ride cars with recalled airbags. People take medicine with wrong dosages, all because the information existed but never reached them.

We built RecallRadar to close that gap.

Simple. Automatic. Potentially life-saving.

Future

We plan to add WhatsApp/SMS alerts, a mobile app, barcode scanning, and expand to global databases like UK, EU, and other countries. A browser extension that auto-adds products when you shop online is also on the roadmap.

Built With

  • apscheduler
  • cpsc-api
  • fastapi
  • fda-api
  • gmail-smtp
  • google-gemini-flash
  • javascript
  • jwt
  • next.js
  • nhtsa-api
  • python
  • railway
  • supabase
  • tailwindcss
  • typescript
  • vercel
Share this project:

Updates