Nelavi

Nelavi: a rover senses soil and spots crop disease on-device, plus a box grades harvested grain — giving farmers real-time data and fair pricing proof, right on their phone.

Comment

Inspiration

Every season, farmers decide how much fertiliser to use mostly by habit or by what the dealer says. Soil tests exist, but they're slow, far away and rare: often one mixed sample per field, years apart. The cost shows up worldwide. About 54% of the nitrogen fertiliser put on crops never reaches the harvest, and FAO's 2026 report says soil health is getting worse in most regions. In India, nitrogen use has drifted to 10.9 : 4.1 : 1 against a recommended 4 : 2 : 1.

We kept asking one question: why can't the soil test come to the field, cost very little, and give advice a farmer can act on that same day? That question became CropSync.

What it does

Nelavi is a low-cost, solar-charged system with three parts:

  • A field rover goes across the farm and checks soil nutrients (N, P, K), moisture and temperature at many points, not just one. It also photographs leaves with a camera on a pan-tilt head.
  • AI turns the readings into a nutrient map of the field and flags leaves that show signs of disease.
  • A phone app tells the farmer what to add, where and how much, in their own language. The rover makes its own Wi-Fi network, so it works in fields with no internet.

A separate grain quality box photographs a grain sample under fixed lighting and estimates its quality, so the farmer knows what the crop is worth before selling.

It's built to be shared: one kit can serve many small farms through a farmer group or co-operative.

How we built it

We split the work three ways:

  • Hardware and design (Muthu): chassis, motors, sensors and wiring.
  • Research and prototyping (Debajyoti): soil science, farmers' needs, and building and testing the prototype.
  • Software and AI (Akshaj): the AI models and the phone app.

The hardware design:

  • Brains: an ESP32 microcontroller runs the rover. An CAM handles leaf and grain photos.
  • Soil sensing:
    • a JXBS-3001 NPK probe, which talks to the ESP32 over RS-485 through a MAX485 chip
    • a capacitive moisture sensor
    • a DS18B20 soil temperature sensor
  • Air and light: a BME280 (air temperature, humidity and pressure) and a BH1750 light sensor.
  • Movement: four 12 V geared motors driven by two BTS7960 motor drivers, plus an HC-SR04 ultrasonic sensor to avoid obstacles.
  • Camera head: a 28BYJ-48 stepper motor turns it left and right, an SG90 servo tilts it, and an LED ring gives steady lighting.
  • Power: a 3-cell 18650 lithium battery pack with a protection board, a 10 W solar panel with a constant-current charger, a 5 V regulator for the electronics, and a 10 A fuse.

We drew the full circuit diagrams and the chassis layout, built a 3D model of the rover, and wrote a complete costed parts list.

On the software side, we planned:

  • a small AI model that checks one crop's leaves for "healthy" or "diseased" first, before trying to name specific diseases
  • a calibration method that compares our probe readings with lab results from the same soil
  • an Android app that connects directly to the rover's Wi-Fi

The first physical prototype is being built now.

Challenges we ran into

  • Pushing the probe into the soil. Our calculations showed the rover can push about 31 N, but firm soil needs 60–120 N. Instead of hiding this, we redesigned the probe as a separate pod on a cable. A person presses it in with their foot, and a depth stop keeps it at 10 cm.
  • Cheap sensors aren't lab-accurate. Low-cost NPK probes give approximate readings, and soil moisture changes them. We designed a calibration step against lab tests. Depending on how well the probe matches the lab, the app shows numbers, simple low/medium/high levels, or tells the farmer to send a sample to a lab.
  • Connecting without internet.A normal web app can't talk directly to the rover's local Wi-Fi because of browser security rules. We switched to an Android app that connects to the rover directly. Battery reality check. Our first runtime estimate was too optimistic. After recalculating, we now say 1–2.5 hours of driving, or 2–4 hours of surveying with stops. -Rice paddies. The rover can't drive in flooded fields. We planned around it: test the soil before planting and after harvest, and check leaves from the bunds or with the phone during flooding.
  • Wiring safety. One of the ESP32's pins briefly switches on when the board starts up, which could make the motors jump. We added pull-down resistors and a single enable line so the motors stay off until the software is ready.
  • Cost. We cut the shopping list from ₹11,161 to ₹9,293 without losing core features.

Accomplishments that we're proud of

  • A complete, checked design: circuit diagrams that pass our automated wiring check, a chassis layout, a 3D model and a costed parts list (about ₹13,600 in total parts, around $150).
  • Finding and fixing real engineering problems on paper, such as probe force, startup glitches, connectivity and battery life, before spending money on parts.
  • Keeping the whole system low-cost, solar-powered and able to work offline, so it suits small farms that big precision-farming tools ignore.
  • Being honest about accuracy: building calibration and confidence levels into the design from day one.
  • Doing all this as first-year students, while learning electronics, AI and agronomy at the same time.

What we learned

  • Real-world physics wins.** A design can look perfect until you calculate forces, current and battery life. Doing the maths early saved us from building the wrong thing.
  • Accuracy and trust matter more than features.** A farmer will only follow advice they trust, so telling them how sure the system is matters as much as the reading itself.
  • Farmers' realities shape the product.** Flooded paddies, no internet, small budgets and local languages all changed our design.
  • Team roles make things move.** Splitting software, hardware and research let us move fast without stepping on each other.
  • The technical skills:** RS-485 sensor communication, motor drivers, battery protection, solar charging, and planning an AI model with limited data.

What's next for CropSync

  1. Finish and test the prototype:** assemble the rover and grain box, then test every sensor on the bench.
  2. Field pilot:** run CropSync on real farms near Woxsen and compare our readings with lab results to prove (or improve) accuracy.
  3. Train the AI:** collect leaf photos for one crop first, build the healthy/diseased model, then add specific diseases.
  4. Build the app:** offline maps, simple advice in multiple languages, and a digital Leaf Colour Chart for nitrogen top-ups in rice.
  5. Partner with farmer groups:** test the shared-kit, pay-per-acre model with a co-operative.
  6. Version 2 hardware:** a motorised soil probe, a weather-sealed body, and a paddy-ready version with wider lug wheels and higher clearance.

Built With

Share this project:

Updates

Submission history