The Vision: Beyond Simple Chatbots
Cabbo represents the evolution of agricultural AI from simple Q&A interfaces to proactive, agentic problem solvers. While many platforms offer a wrapper around an LLM, Cabbo was engineered as a comprehensive Decision Support System (DSS) that actively researches, analyzes, and reasons about agricultural problems.
It isn't just about answering questions; it's about performing work.
The "Deep Research" Engine
At the heart of Cabbo is a sophisticated Deep Research Agent, designed to autonomously gather high-quality intelligence from the web. We realized that for complex agronomic queries, a single search is never enough.
Adaptive Research Strategies
The system implements an Adaptive Strategy pattern (ResearchStrategyHandler) that analyzes the complexity of a user's request and dynamically selects the best approach:
- Quick Strategy: For simple fact-checking, it performs a rapid, targeted search using top-3 sources.
- Comprehensive Strategy: For complex problem-solving, it initiates a multi-stage workflow: finding sources, extracting key facts, cross-referencing information, and synthesizing a final report.
- Academic Strategy: A specialized mode that prioritizes
.edu,.gov, and research repositories, filtering out commercial noise to focus on peer-reviewed agronomic science.
Source Quality Scoring
To ensure the advice given to farmers is reliable, we implemented a rigorous Source Quality Scoring algorithm. The system doesn't just read any link; it evaluates every URL based on:
- Authority Score: Heavily weighted towards trusted domains (e.g.,
pubmed.ncbi.nlm.nih.gov,extension.university.edu). - Relevance Score: Calculated via semantic keyword overlap between the query and the page title.
- Content Type: It distinguishes between "Academic", "Government", "News", and "Commercial" content, prioritizing unbiased information.
Only sources that pass a configurable Min Confidence Threshold (e.g., 0.7) are used in the final synthesis.
Robust Architecture
Cabbo is built on a resilient Microservices Architecture powered by FastAPI and Docker.
Captcha-Resistant Browsing
One of the biggest challenges in automated research is access. Our pipeline utilizes BrowserUse to drive actual browser instances, allowing the agent to navigate complex websites just like a human. To maintain high availability, we implemented a robust fallback mechanism:
- Primary: SearXNG instances (hosted internally) for aggregation without tracking.
- Secondary: Direct website navigation if aggregators fail.
- Protocols: Strict "No-Direct-Google" policies to avoid IP bans and Captcha traps.
User Experience: Field-Ready Design
This powerful backend is exposed through a mobile-first interface designed for the field. It supports multi-modal input, allowing farmers to upload images of crops directly for analysis relative to the research findings. The system preserves session context, allowing for long-running investigations ("How is that disease we discussed last week progressing?").
Figure 1: The Cabbo mobile-first interface, designed for clarity and ease of use in outdoor conditions.
