Agrinput is a mobile application designed for the agricultural sector in Italy. It enables users—such as farmers, agronomists, and technicians—to scan product packaging and instantly receive official technical information about agrochemicals and fertilizers, enhancing decision-making in the field.
Developed as a complete solution integrating OCR, AI, and official ministerial databases, Agrinput addresses the challenge of identifying products in real-world conditions, where product names are often unclear or inconsistently printed on packaging.
Key Highlights
- Scans product labels via camera and extracts meaningful data using OCR + AI.
- Fetches validated information from Fitogest (agrochemicals) and Fertilgest (fertilizers).
- Supports multi-result output with manual selection, ideal for ambiguous or low-quality inputs.
- Stores local scan history and lets users create custom collections for frequent reference.
Technical Workflow
-
Image Preprocessing
- Improves image quality for optimal OCR performance.
- Handles real-world camera inputs with filters and adjustments.
-
Text Extraction
- Uses Tesseract OCR, containerized via Docker.
- Extracts unstructured text from the image for semantic analysis.
-
AI Keyword Interpretation
- Utilizes OpenAI GPT to identify keywords or registration codes from raw text.
- Enhances accuracy in searching for relevant products, even with noisy input.
-
Database Search
- Matches extracted data with official registries (Fitogest, Fertilgest).
- Displays detailed technical sheets, safety info, and update history.
-
User Result Selection
- Users manually select the correct product from AI-generated matches.
- Ensures flexibility and retrocompatibility with previous app versions.
Tech Stack
Frontend (Mobile)
- React Native + Expo Go
- NativeWind (Tailwind-style utility classes)
- AsyncStorage for local persistence
- TypeScript
- Deep Linking
Backend
- FastAPI (Python 3.11)
- Tesseract OCR
- PostgreSQL
- Redis (for caching)
- OpenAI GPT (semantic keyword extraction)
DevOps
- Docker + Portainer (container orchestration)
- GitHub Actions (CI/CD pipeline)
- AWS ECR (Docker image registry)
- Nginx (reverse proxy)
Role & Responsibilities
As the Lead Developer, I was responsible for the end-to-end architecture and implementation of the project:
- Designed the full mobile-to-backend OCR pipeline, integrating AI for intelligent product matching.
- Developed a high-performance API in FastAPI with PostgreSQL, optimized for rapid queries and secure access.
- Architected and deployed services using Docker, managed via Portainer, with automated updates via GitHub Actions.
- Created the mobile frontend using React Native, focusing on usability, offline support, and responsive UI with NativeWind.
- Collaborated with the design team to ensure a clear and intuitive user experience aligned with field requirements.
Impact
- Reduced the time and effort for agricultural professionals to access compliant product data.
- Improved reliability of product identification using AI and OCR in tandem.
- Enabled informed decision-making directly from the field with mobile-first usability.
Screenshots
Agrinput represents the fusion of AI, OCR, and mobile usability to solve real challenges in the agricultural industry.