Introduction
OFELIA (Offline Forage Expert and Life Identification Assistant) is a multilingual mobile application that enables offline recognition of fungi, plants, and animals in nature.
It was born from a simple observation: in the field, internet access is often unavailable, yet correct species identification can be vital — for safety, research, and sustainable resource use.
By combining on-device artificial intelligence, real-time image recognition, and a conversational assistant, OFELIA empowers hikers, foragers, researchers, and field professionals to make informed decisions anytime, anywhere.
The project originated in Hungary and is being developed in collaboration with professional partners, including the Hungarian Mycological Society and the OpenBioMaps research community. The long-term vision is for OFELIA to become a globally recognized field decision-support assistant.
Problem Statement
Every year, hundreds of poisoning incidents occur due to the misidentification of fungi and plants — most of which could be prevented with accurate identification tools. Many species are visually similar, and even experts rely on subtle morphological traits.
Current mobile applications either depend on a continuous internet connection or lack the precision required for reliable identification in the wild. In field, military, or emergency contexts, speed and accuracy are often matters of safety.
Furthermore, non-experts need clear, human-readable guidance, while researchers require tools capable of collecting structured, interoperable data compatible with existing ecological databases. No comprehensive offline solution currently bridges these user groups — a gap that OFELIA is designed to fill.
Beyond research and civilian applications, the lack of reliable offline recognition tools represents a strategic vulnerability. In emergency or crisis situations, access to an autonomous field identification system could support public safety, environmental monitoring, and national resilience.
We believe such a system should ultimately be considered part of the national civil protection and defense infrastructure, ensuring availability to citizens and professionals even under disrupted communication conditions.
Objectives
-
Deliver accurate offline species identification in remote environments.
-
Provide modular model packages, allowing users to choose which taxa to download.
-
Ensure real-time results with optimized AI inference.
-
Support citizen safety, biodiversity research, and field operations in critical environments.
-
Integrate species data with OpenBioMaps and other open ecological databases.
-
Develop towards LLM-based intelligent agents for contextual interaction and knowledge reasoning.
-
Contribute to environmental education and sustainable forest management.
-
Establish OFELIA as a component of the national civil protection and defense infrastructure, ensuring that in emergencies or network disruptions, citizens and field operators retain access to accurate species identification and ecological data.
Solution
OFELIA is a hybrid AI-driven system integrating advanced recognition models with an intuitive, multilingual mobile interface, ensuring accessibility for a broad international user base.
The app enables users to photograph a fungus, plant, or animal and receive an identification result within seconds, even without connectivity. A forthcoming chat-based assistant will allow users to ask follow-up questions and receive contextual explanations.
Unlike conventional classifiers, OFELIA employs a human-in-the-loop approach: the AI may ask targeted questions (e.g. spore print color, gill attachment, stipe shape) to refine its decision. This interactive exchange merges human intuition with machine precision, greatly improving reliability.
Offline functionality ensures full usability in remote, data-restricted, or mission-critical scenarios.
Technology
OFELIA’s architecture is modular, scalable, and serverless, ensuring privacy and resilience.
All AI inference runs on-device, with no user data transmitted to external servers.
The current prototype employs ONNX-based computer vision models, including a YOLOv11n-CLS classifier capable of predicting species in 40–45 ms on an iPhone 15 Pro (and under 2 ms in the native implementation).
During Phase 1, OFELIA uses a serverless data flow, where user observations are transferred directly to the OpenBioMaps system, ensuring compatibility with national biodiversity databases. The Conservation Officer application accesses aggregated datasets from OpenBioMaps to support environmental monitoring and data-driven conservation decisions.
Following the initial PWA prototype, native iOS and Android applications will already be delivered in Phase 1, featuring enhanced performance, offline data management, and tight integration with device sensors and AI inference engines.
The architecture is designed to support future LLM integration, enabling agentic AI functionality such as reasoning, dialogue, and task automation.
From its inception, OFELIA has been built for scalable expansion with new species groups and modules. Energy efficiency is a core design principle — models are optimized through quantization and accelerated inference to minimize power consumption while preserving accuracy.
Open Science & Collaboration
All training datasets and AI models are based on openly licensed (CC-BY / CC-BY-SA) sources, ensuring transparency, reproducibility, and ethical data use.
OFELIA actively contributes to the open science ecosystem, promoting cross-border collaboration and citizen participation in biodiversity monitoring. The system is designed to interoperate with public APIs and open data infrastructures such as OpenBioMaps, GBIF, and iNaturalist.
Partnerships with scientific and conservation organizations ensure both ecological accuracy and real-world applicability.
