DEV Community

Ronak Mahajan
Ronak Mahajan

Posted on

I'm 13 and Building an AI Produce Scanner for Local Markets: Meet Sprout Atlas

#ai #buildinpublic #shipaton #webdev

Hey DEV Community! đź‘‹

I'm Ronak, a 13-year-old student builder from Pune, India.

Over the past few weeks, I’ve been building Sprout Atlas—an interactive produce field guide and AI-powered camera scanner designed to solve a very real, everyday problem at local markets.

Here is the story behind why I’m building it, the architecture under the hood, and what I’ve learned so far building in public for RevenueCat’s #Shipaton 2026.

đź›’ The Problem: Buying Blind at the Mandi

In India and many parts of the world, millions of families buy fresh produce at local markets (mandis) every single week. But when you buy from a stall, zero information changes hands beyond price and weight.

Shoppers are constantly left wondering:

  • Is this fruit actually fresh, or was it harvested days ago?
  • Is this glossy apple coated in synthetic wax?
  • Were these bananas force-ripened with chemicals like calcium carbide?
  • How do I safely rinse away pesticide residues before cooking?

Because of this uncertainty, most people fall into what I call the "4-Item Diet Trap": sticking strictly to Apple, Banana, Tomato, and Potato.

Dozens of rich, seasonal local vegetables sit right on the cart, but go unbought because search engines fail when you don't even know what name to type into Google.

🌿 What Sprout Atlas Does

Sprout Atlas meets the shopper directly at the point of purchase:

  1. 500+ Botanical Produce Guides: Comprehensive breakdowns of familiar and rare produce—what parts are edible, seasonality, taste profiles, and cooking preparation.
  2. Multi-Lingual Alias Index: In India, a single vegetable can have 5+ regional names (e.g., Suran / Jimikand / Elephant Foot Yam). The database maps phonetic regional names across Marathi, Hindi, Tamil, and Bengali.
  3. Science-Backed Pesticide Rinsing Guides: Plain tap water removes less than 20% of hydrophobic chemical residues. We provide simple, step-by-step soaking protocols (e.g., 1 tsp baking soda in cold water for 3–5 minutes) to degrade surface residues by up to 90%.
  4. Point-of-Purchase AI Camera Scanner: Point your phone camera at produce to assess visible freshness and flag artificial ripening signals.
  5. Interactive Labs: Includes a conversational AI Nutrition Tutor and a daily "Rainbow Challenge" tracker to encourage diverse eating.

đź’» Tech Stack & Architecture

Here is the stack and workflow I'm using to build Sprout Atlas:

  • Design & UI System: Built with a custom "Field Guide" design system—warm vintage paper backdrops (#FAF6E9), deep forest ink (#233022), responsive 9:16 mobile frames, and hand-drawn SVG vector doodles for produce items.
  • Prototyping & Code: Semantic HTML5, CSS Grid/Flexbox, and vanilla JavaScript for snappy, lightweight performance.
  • AI & Computer Vision: Google AI Studio / Gemini API for multimodal image classification, ripeness detection, and conversational nutrition tutoring.
  • Monetization Engine: Integrating the RevenueCat SDK to handle freemium entitlement logic (3 free AI scans/week + unlimited Pro tier).

🛠️ The Biggest Challenges So Far

Building this as a 13-year-old solo dev has been an intense learning curve. The top three problems I've faced:

  1. One-Handed Market UX: Nobody at a vegetable stall has two hands free—you’re holding a shopping bag in one hand. We had to ruthlessly eliminate buttons so the entire scanning and safety check takes 1 tap and under 2 seconds.
  2. Cosmetic Blemishes vs. Real Spoilage: In organic local markets, surface spots often mean less pesticide spray, not bad fruit! Training our AI vision logic to distinguish harmless cosmetic scars from actual microbial rot was crucial to avoid causing food waste.
  3. Fighting AI Code Bloat: AI coding assistants generate boilerplate fast, but they love adding 10 unnecessary buttons and bloated state. I've learned that developer taste and subtraction matter far more than prompt speed.

🚀 What’s Next?

I’m participating in RevenueCat's #Shipaton 2026 under the NextGen and #BuildInPublic tracks. Over the next few weeks, I’ll be sharing our live camera testing results, database benchmarks, and paywall experiments.

I’d love your feedback!

If you were building an on-device camera scanner for messy, outdoor lighting environments:

  • What client framework would you choose (Flutter vs. React Native vs. Swift)?
  • How would you optimize image compression before hitting the vision model?

Let me know in the comments below! 👇

Follow my build journey on Twitter/X: [@RonakMahajan13] | Project: Sprout Atlas 🌱

Top comments (0)