Olivision (زيتونك)

uses models/gemini-3-flash-preview to predict olive oil quality from a photo. We helping farmers harvest at the perfect moment to maximize "Extra Virgin" profits

Comment

Inspiration

The inspiration for Olivision (زيتونك) comes from the Harvest Gamble that millions of olive farmers face every season. In the Mediterranean, olive oil is more than just a product; it’s a livelihood.

I witnessed how farmers struggle to decide the exact day for harvesting. A mistake of just one week can lead to increased acidity and the loss of the Extra Virgin certification—cutting profits by up to 30%.

Olivision aims to empower small-scale farmers with a digital agronomist that turns a simple smartphone photo into a sophisticated chemical analysis.

What It Does

Extra Virgin is an AI-powered co-pilot that predicts the quality and acidity of olive oil before the olives are harvested.

By simply taking a photo of an olive cluster, the app:

  • Analyzes the Maturity Index (MI) based on skin and pulp pigmentation
  • Cross-references visual data with local climate history (temperature & humidity)
  • Provides a Harvest Window recommendation to ensure premium quality
  • Estimates the potential acidity level and flavor profile (bitterness / pungency)

How We Built It

The project is built on the models/gemini-3-flash-preview model, leveraging its native multimodal capabilities to process both images and structured environmental data simultaneously.

Vision Analysis

We use Gemini to perform zero-shot classification of olives into 7 ripeness categories.

Mathematical Core

The app applies the standard Maturity Index (MI) formula:

$$MI = \frac{\sum_{i=0}^{7} (i \times n_i)}{N}$$

Where:

  • (i) is the color category (0 = green, 7 = black)
  • (n_i) is the number of olives in that category
  • (N) is the total number of olives

Reasoning Engine

Gemini correlates the computed (MI) with Phenolic Content curves embedded in the system prompt to predict the oil’s stability and commercial grade.

Environmental Variability

Shadows, leaf density, and varying sunlight can easily skew color detection. Simple RGB analysis proved insufficient.

We overcame this by prompting Gemini to reason about lighting conditions and normalize pigmentation classes accordingly.

Cultivar Diversity

Olive varieties such as Chemlali, Picual, and Arbequina mature differently.

We addressed this by allowing Gemini to identify the cultivar from leaf shape and fruit morphology before calculating ripeness.

Accomplishments We’re Proud Of

  • Zero-Hardware Solution: Replicated results of expensive laboratory NIR sensors using only a smartphone camera and AI
  • Scientific Grounding: Rooted in established agronomical formulas ((MI)) rather than black-box predictions
  • Accessibility: Designed for traditional farmers with zero technical background

What We Learned

During the study phase, we learned that multimodality is the true key to Smart Agriculture.

An image alone is just pixels—but when combined with 10-day weather forecasts and botanical data, it becomes a powerful predictive tool.

We also learned the importance of edge-case handling, such as detecting diseased fruits that could compromise an entire oil batch during pressing.

What’s Next for Olivision (زيتونك)

  • Hyper-local Weather Data: Integrating local APIs to monitor Accumulated Thermal Units (ATU) per grove
  • The Pressing Guide: Recommending optimal extraction temperature (T < 27^{\circ}C) for cold pressing
  • Marketplace Integration: Connecting farmers with Premium Extra Virgin predictions directly to buyers and exporters

link demo : https://olivision.vercel.app/

Built With

Share this project:

Updates