-
-
interface
Inspiration
Traditional soil test summaries are often dense and full of scientific metrics that smallholder farmers struggle to convert into immediate actions. We wanted to build an autonomous agent that bridges this gap, interpreting informal observations and raw metrics in real time to optimize regional crop production.
What it does
AgroSoil-Qwen acts as an autonomous digital agronomist. It parses unstructured natural language inputs regarding field conditions (such as low phosphorus findings), extracts raw values, and routes them to precise Python calculation tools. It returns an optimized fertilizer remediation plan alongside a "Human-in-the-Loop" verification checkpoint for enterprise safety.
How we built it
The intelligence backend utilizes the flagship Qwen-Max model on Qwen Cloud, integrated via the international OpenAI-compatible DashScope API endpoint. The data processing loop leverages native tool-calling to offload deterministic mathematical computations to local Python modules, preventing LLM hallucinations. The frontend is engineered using a clean Streamlit dashboard.
Challenges we ran into
We encountered local environment routing barriers and network socket drops during early development on Windows PowerShell. We resolved these infrastructure challenges by scaling up the client configuration with dedicated execution limits (timeout=60.0) and automated network packet recovery settings (max_retries=5) to ensure a resilient production environment.
Accomplishments that we're proud of
We are proud of engineering a highly accurate tool-calling loop that seamlessly connects an advanced cloud LLM with local mathematical verification code. We successfully preserved system guardrails by introducing a functional human validation step prior to automated procurement handoffs.
What we learned
We gained deep insight into enterprise workflow automation using Qwen Cloud APIs, optimal prompt orchestration for precise json-schema tool calling, and cross-platform environment variable management under Windows architectures.
What's next for AgroSoil-Qwen-Autopilot
We intend to integrate multi-modal image evaluation to allow farmers to upload photos of crop nutrient leaves directly alongside soil data, alongside extending the database schema to handle larger regional soil fertility mapping.
Log in or sign up for Devpost to join the conversation.