Buzy AI is a LoRA adapter built on Gemma 4 that specializes the model for explainable business reasoning over enterprise documents.
Rather than acting as a generic chatbot, Buzy AI transforms business information into structured, evidence-based recommendations to support decision-making.
Model Details
Developed by: Destin Biringanine
Entreprise: Code, Growth Alive
Base model:unsloth/gemma-4-E2B-it-unsloth-bnb-4bit
Fine-tuning: LoRA (PEFT)
Framework: Unsloth
License: Apache-2.0
Intended Use
This model is designed for enterprise decision-support tasks, including:
Contract analysis
Invoice understanding
Procurement intelligence
Supplier risk assessment
Financial document analysis
Business reporting
Explainable recommendations
The model follows a structured reasoning process consisting of:
Evidence
Reasoning
Confidence
Business Impact
Recommended Actions
Training
The adapter was fine-tuned using Unsloth on a structured business reasoning dataset.
The objective was not to teach new business knowledge, but to improve the model's ability to produce consistent, transparent, and explainable business reasoning.
Example
Input
Supplier A has delayed deliveries for three consecutive months.
The defect rate increased by 18%.
The contract expires next month.
Should the contract be renewed?
Output
json
1{2"decision":"Do not renew without corrective actions.",3"evidence":[4"Three consecutive delivery delays",5"18% increase in defect rate"6],7"reasoning":"Operational performance has consistently deteriorated, increasing supply-chain risk.",8"confidence":0.95,9"business_impact":"High operational risk",10"recommended_actions":[11"Request a corrective action plan",12"Evaluate alternative suppliers",13"Conduct a supplier performance review"14]15}
Limitations
This model is intended for research and demonstration purposes.
It should be used as a decision-support assistant rather than an autonomous decision-maker and performs best when combined with an external knowledge base or retrieval system.
Acknowledgements
This project was developed for the Build with Google Gemma Hackathon using: