AgroPulse AI

Inspiration

AgroPulse AI started with a simple question: what if a poultry farmer could have an intelligent assistant that understands not just a disease, but what is actually happening on the farm?

Farmers have to keep track of many things at once — their birds, mortality, feed, medication, vaccinations and egg production. A lot of this can still be handled manually, and when internet access is unreliable, depending on cloud-based AI is not always practical. I wanted to build something that could bring these things together while still working offline.

What started as an idea for an AI poultry assistant gradually became something much bigger.

What it does

AgroPulse AI is an offline-first AI-powered farm intelligence and smart farm operations platform for poultry farmers.

Farmers can manage livestock, mortality, feed inventory, medication, vaccinations and egg production in one place. The system uses these records to provide useful insights such as farm risk, biosecurity scores, disease trends, possible disease clusters and farm health history.

It also has Smart Farm Operations that help farmers make daily decisions. AgroPulse can calculate feed days remaining and reorder planning, identify mortality trends, compare egg production with previous averages, track vaccination status, calculate a farm performance score and generate recommendations based on the farm's actual records.

The AI assistant can also help farmers describe symptoms and receive an AI-assisted assessment using a local veterinary knowledge base and a locally running language model. The system is designed to support farmers and encourage veterinary attention when necessary, rather than replace a veterinarian.

How we built it

I built AgroPulse AI with Python and Streamlit, using SQLite for farm records and ChromaDB for local knowledge retrieval. The AI pipeline uses a local veterinary knowledge base, local embeddings, RAG, and Phi-3 Mini running locally through Ollama.

I also added local speech recognition and text-to-speech so the platform can support voice interaction. The analytics and Smart Farm Operations features are mostly based on transparent calculations and rules using actual farm records rather than asking the AI model to guess.

The project was developed gradually through multiple sprints. I started with the core AI consultation pipeline, then added risk analysis and voice interaction, followed by farm intelligence, farm management, analytics and finally Smart Farm Operations.

A major goal throughout the development was keeping the system offline-first, so the core features do not depend on sending farm data to a cloud AI service.

Challenges we ran into

One of the biggest challenges was connecting many different parts of the system without breaking the features that had already been built. The project grew from a simple AI assistant into a system containing AI, databases, analytics, voice interaction, farm management and recommendation features.

Running a local language model on a laptop also came with performance and memory limitations. I had to pay attention to how much processing was happening and avoid unnecessary operations.

Another challenge was building the interface with Streamlit. Some navigation approaches that looked good initially did not work reliably, so I had to test different approaches and adapt the design around what worked consistently.

Accomplishments that we're proud of

I am most proud of how much AgroPulse evolved from the original idea. It is no longer just a chatbot that answers questions about poultry diseases. It combines AI with actual farm records to give farmers a clearer picture of what is happening on their farm.

I am also proud that the system can work offline, including the local AI, knowledge retrieval and voice features. The Smart Farm Operations system is another part I am particularly proud of because it turns farm records into practical information such as feed planning, mortality trends, egg production insights and daily recommendations.

Most importantly, I built the project around transparency. When AgroPulse makes a recommendation, it should be possible to understand why that recommendation appeared and what farm data triggered it.

What we learned

This project taught me that building a useful AI product is not just about putting an AI model into an application.

I learned a lot about RAG, local LLMs, GGUF models, SQLite, ChromaDB, speech recognition, data analysis and Streamlit application development. I also learned how important it is to build gradually, test each part and think about how different features work together.

The biggest lesson for me was knowing when to use AI and when not to. Some problems are better solved with simple calculations and transparent rules. Combining those approaches with AI made AgroPulse more practical and easier to understand.

What's next for AgroPulse AI

The next step is to continue improving AgroPulse for real farm use. I want to expand its disease knowledge, improve local-language voice support and make its farm forecasting and recommendations more useful as more farm data is collected.

I also see opportunities to extend AgroPulse beyond poultry farming and eventually connect it with sensors and other sources of agricultural data.

The long-term goal is to build AgroPulse into a practical offline-first farm intelligence platform that helps farmers make better decisions every day, even when reliable internet access cannot be taken for granted.

Built With

• ai
• chromadb
• gguf
• llm
• machine-learning
• offline-ai
• ollama
• pandas
• phi-3
• plotly
• python
• rag
• speech-recognition
• sqlite
• streamlit