BAU AgriSmart AI Platform

Inspiration

Farmers and agricultural extension officers in Bangladesh face critical delays in diagnosing crop diseases, pests, and nutrient deficiencies. Language barriers and complex technical guidelines often prevent smallholder farmers from accessing timely help. Inspired by the research community at Bangladesh Agricultural University (BAU), we built BAU AgriSmart AI to bring institutional agronomic knowledge directly to farmers in simple, bilingual terms.

What it does

BAU AgriSmart AI is a real-time, stateless decision-support platform designed for Bangladesh's agricultural sector:

• 💬 AI Agronomist Consultation: A bilingual chatbot providing diagnosis, chemical/organic treatment plans, and prevention advice for paddy, vegetables, fruits, and poultry.
• 📸 Visual Leaf Disease Scanner: Multimodal AI diagnosis that analyzes uploaded photos of infected crops to identify pathogens instantly.
• 🧮 Smart Field Fertilizer Engine: Calculates precise dosages of Urea, TSP, MoP, and Gypsum based on land area in decimals and crop type.
• 🔊 Audio Accessibility: Generates real-time speech output in Bangla and English for farmers who prefer audio guidance.

How we built it

• Frontend: Built using Streamlit with dynamic CSS styling and custom state management for language toggling.
• Core Engine: Powered by Google's Gemini API (gemini-3.6-flash) via the google-genai SDK for multimodal vision and text inference.
• Audio Synthesis: Integrated gTTS (Google Text-to-Speech) to stream audio responses dynamically.
• Image Processing: Implemented using Pillow to capture and format leaf pathology uploads.

Challenges we ran into

• Bilingual UI State Management: Ensuring that switching between English and Bangla smoothly updated every interface label, button, and prompt structure without resetting user input states.
• Cloud Dependency Alignment: Resolving deployment package builds on Streamlit Cloud by structuring exact dependencies in requirements.txt.
• Prompt Engineering for Local Context: Fine-tuning system prompts to ensure disease treatments align with fertilizer availability and agricultural practices in Bangladesh.

Accomplishments that we're proud of

• Delivering a zero-latency, highly accessible user experience that works across mobile web browsers.
• Successfully implementing dual-language support (English & Bangla) across both the visual interface and synthesized voice outputs.
• Building a full-stack, multimodal solution with zero reliance on complex traditional database overheads.

What we learned

• Harnessing Gemini's gemini-3.6-flash model for lightweight, high-speed multimodal reasoning.
• Structuring conversational prompt pipelines to enforce localized, safety-conscious agricultural outputs.
• Designing accessible UI/UX tailored for non-technical agricultural workers.

What's next for BAU AgriSmart AI Platform

• Localized Weather & Spray Warnings: Integrating real-time weather APIs to warn farmers against applying pesticides before rain.
• BAU Outbreak Heatmap: Adding a spatial database to track crop disease clusters across different divisions in Bangladesh.
• Offline SMS / Voice Gateway: Expanding access to feature phones via automated IVR calls and SMS alerts.

Built With

• cloud
• gemini-3.6-flash
• github
• gtts
• pillow
• python
• streamlit