KishanMitra

Hindi-first AI crop disease detector for 140M Indian farmers. Built entirely on MeDo — zero code, full impact. 🌾

Comment

KisanMitra AI

Inspiration

It started with a number: ₹50,000 crore.

That's how much Indian farmers lose every year to crop diseases — not because cures don't exist, but because the knowledge never reaches them in time.

India has 140 million farmers. Most of them speak Hindi. Most of them own a smartphone. But every crop disease detection tool ever built assumes they speak English, have internet bandwidth, and can afford an agronomist.

We thought about a farmer in rural Bihar — watching his tomato crop turn brown, not knowing why, with the nearest agriculture officer 40 kilometers away. By the time he gets an answer, the crop is gone. The season is lost. The loan isn't repaid.

That farmer deserved better. That's why we built KisanMitra AI.

The name means "Farmer's Friend" in Hindi. That's exactly what we set out to build — not an app, but a friend that speaks your language, understands your problem, and tells you exactly what to do.

What it does

KisanMitra AI is India's first Hindi-first, AI-powered crop disease detection platform built entirely on MeDo.

A farmer opens the app and takes a photo of their sick crop. Within seconds, they receive:

  • 🌿 Disease name — in Hindi with English in brackets
  • ⚠️ Severity level — Mild, Moderate, or Severe with color coding
  • 🔬 Visual overlay — disease area highlighted directly on their photo
  • 💊 Medicine recommendation — with exact brand name and dosage per liter
  • 🌱 Organic alternative — neem oil or other accessible local remedies
  • 📉 Yield loss estimate — so they understand urgency
  • 🗣️ Hindi voice readout — for farmers who struggle to read

Beyond diagnosis, KisanMitra AI offers:

  • AI Chatbot — answers farming questions in Hindi (fertilizers, govt schemes, weather effects)
  • Diagnosis History — saves past scans so farmers track disease patterns
  • Learning Mode — teaches about 4 major Indian crop diseases interactively
  • Language Toggle — switches between Hindi and English instantly
  • Pricing Plans — Free (3 scans/day), Farmer Pro (₹99/mo), Agriculture Expert (₹299/mo)

How we built it

"We didn't write code. We had a conversation. MeDo wrote the code."

The Stack (All Generated by MeDo)

Layer Technology
Frontend React + Tailwind CSS
Animations Framer Motion
AI Vision Claude Vision API
Voice Browser SpeechSynthesis (Hindi)
Disease Overlay HTML5 Canvas
Database Supabase (via MeDo Plugin)
Image Upload ImgBB API (via MeDo Plugin)
Location ip-api (via MeDo Plugin)
Deployment MeDo one-click deploy

The MeDo Build Process

Turn 1 — Foundation We described the entire app in one Deep Build prompt: dark cinematic UI, image upload, AI diagnosis pipeline, Hindi output, mobile-first layout. MeDo generated the complete frontend structure.

Turn 2 — Image Fix The initial image upload had a size issue with large photos. We asked MeDo to add Canvas-based compression (resize to 800×800, JPEG quality 0.7) before the API call. Fixed in one turn.

Turn 3 — Feature Bundle In a single prompt, we asked MeDo to add: authentication (login/signup), language toggle (Hindi/English), floating AI chatbot, feedback system after each diagnosis, and pricing modal. MeDo added all five features without breaking existing functionality.

Turn 4 — Plugin Integration We connected three custom plugins:

  • ImgBB — for image hosting
  • Supabase — for diagnosis history storage
  • ip-api — for geolocation to show nearest Krishi Vigyan Kendra

Turn 5 — Polish Final UI refinements: neon scan animation over uploaded images, Hindi voice readout using window.speechSynthesis, severity color system (green/amber/red), and mobile responsive fixes.

Total turns used: ~12 Total credits used: ~180 Time taken: ~6 hours

What would have taken a team of 3 developers 2-3 weeks — MeDo built in a single day.

The AI Diagnosis Pipeline

$$\text{Image} \xrightarrow{\text{Canvas Compress}} \text{Base64} \xrightarrow{\text{Claude Vision}} \text{JSON} \xrightarrow{\text{Parser}} \text{Hindi UI}$$

The Claude Vision API receives the compressed image with a structured system prompt that forces output as valid JSON — disease name, severity, affected part, treatment steps, medicine with dosage, organic option, yield loss estimate, and a short voice_summary field used by the Hindi TTS engine.

Challenges we ran into

1. Image Size Rejection

The biggest technical blocker. Claude Vision API rejected large mobile photos (>4MB).

Solution: We asked MeDo to implement Canvas-based client-side compression before the API call. Images are now resized to 800×800px at JPEG quality 0.7 — always under 500KB.

2. Hindi Font Rendering

Devanagari script wasn't rendering cleanly on the dark background.

Solution: MeDo added Noto Sans Devanagari from Google Fonts with proper font-display: swap — crisp Hindi text on all devices.

3. Plugin Configuration

MeDo's custom plugin system required precise API descriptions. The Supabase plugin needed exact endpoint structure, auth headers, and schema definition.

Solution: We wrote detailed API descriptions for each plugin (ImgBB, Supabase, ip-api) that MeDo could parse and configure correctly.

4. Keeping Features from Breaking Each Other

Adding auth, chatbot, language toggle, and feedback in sequence risked breaking the core diagnosis flow.

Solution: We bundled all five features into one single MeDo prompt with an explicit instruction: "Do everything in one single build — do not break existing diagnosis functionality." MeDo handled the integration cleanly.

5. Voice in Hindi

window.speechSynthesis works differently across browsers. Android Chrome supports hi-IN well but some desktop browsers default to English.

Solution: We added a language fallback — if hi-IN voice isn't available, it falls back to the first available voice with a slower rate (0.85) for clearer pronunciation.

Accomplishments that we're proud of

🏆 Zero lines of code written manually — the entire platform was built through MeDo conversation turns.

🌾 Real problem, real solution — not a demo toy. This is a tool that could genuinely help a farmer in Bihar save their harvest.

🗣️ Hindi voice output — we're proud that a farmer who struggles to read can still get a full diagnosis explained to them out loud in their language.

📱 Mobile-first — every feature works on a ₹8,000 Android phone, which is the real device of our target user.

🔌 Three plugins integrated — image hosting, database, and geolocation working together seamlessly — all configured through MeDo without touching any config files manually.

⚡ Full-stack in 6 hours — auth + AI vision + chatbot + history + pricing + deployment. The speed MeDo enabled is genuinely remarkable.

What we learned

1. MeDo rewards specificity. Vague prompts give average results. When we described exact colors (#00ff88), exact behaviors ("compress before API call"), and exact edge cases ("if not a crop, return is_crop: false") — MeDo delivered precisely.

2. Bundle, don't stack. Adding features one by one caused integration conflicts. Bundling related features into a single turn with clear separation was far more effective.

3. Plugin descriptions are everything. MeDo's plugin parser is powerful but needs structured input. A well-written API description (endpoint, method, auth, parameters, response schema) gets a working plugin in one shot.

4. No-code doesn't mean no-thinking. The hardest part wasn't getting MeDo to write code — it was knowing what to ask for. Product thinking, UX decisions, and system design still required human expertise. MeDo handled execution; we handled vision.

5. Impact is the best differentiator. In a hackathon full of productivity tools and games, a platform built for 140 million farmers stands out — not because of technology, but because of purpose.

What's next for KisanMitra

Short Term (1 month)

  • [ ] WhatsApp Integration — share diagnosis report directly to farmer's WhatsApp
  • [ ] Offline Mode — cache top 50 disease diagnoses for zero-connectivity areas
  • [ ] Regional Languages — Bhojpuri, Marathi, Telugu, Punjabi support
  • [ ] PDF Export — printable diagnosis report for KVK officers

Medium Term (3 months)

  • [ ] Disease Heatmap — show which diseases are spreading in which districts (aggregated anonymous data)
  • [ ] Government Scheme Integration — auto-match farmer's crop + location to eligible PM schemes
  • [ ] Expert Connect — book 15-min video call with a real agronomist at ₹49
  • [ ] Weather-based Alerts — "High humidity detected in your area — watch for fungal diseases"

Long Term (12 months)

  • [ ] KVK Partnership — official integration with India's 731 Krishi Vigyan Kendras
  • [ ] IoT Sensor Support — connect soil sensors for predictive disease alerts before symptoms appear
  • [ ] Farmer Community — regional language forums where farmers share diagnosis photos and solutions
  • [ ] Insurance Integration — AI-verified disease reports for faster crop insurance claims (PMFBY)

$$\text{1 farmer helped} \rightarrow \text{1 family fed} \rightarrow \text{1 village stronger}$$

KisanMitra AI is not just an app. It is infrastructure for the farmers who feed a billion people.

Built with MeDo. Built for Bharat. 🌾

Submitted to MeDo Hackathon — "Surprise Us!" Track

Built With

Share this project:

Updates