Automated classification to specialized veterinarians based on an NLP language model

The client faced an inefficient process of assigning animal patients to specialized veterinary doctors. Manual decision-making led to delays, inaccuracies, and customer dissatisfaction. Automating the decision process was therefore critical to increasing efficiency, reducing costs, and ensuring better customer experience.

An NLP classification model based on FastText was developed, which recommends the appropriate specialist based on veterinary reports. The model processes diagnostic descriptions, leverages word embeddings, and can accurately predict specialization. The solution runs in real time (approx. 0.03 s/100 requests) on a standard CPU, significantly reducing costs and ensuring easy scalability.

Reception and call center staff (first customer contact).
Veterinary assistants (preliminary triage).
Department heads (capacity and resource allocation).

The total implementation took 3 months (2 months for model development and testing, 1 month for integration).

FastText, Python, BERT NLP Model, Azure Container Services, JSON API, basic NLP preprocessing (tokenization, bigrams, stopword filtering).

Head of IT / Data / Technology

Full implementation

Joint management – model runs on the client’s Azure infrastructure, with technical support provided by the vendor.

IT administrator (deployment and monitoring).
Data specialist (new data preparation, retraining).
End users (regular interface usage).

Low requirements: runs on a single CPU server, basic cloud hosting (Azure). No GPU required, minimal operational costs.

• Faster and consistent specialist recommendations.
  
• Higher client satisfaction and improved customer experience.
  
• Reduced decision-making errors.
  
• Support for data-driven management decisions.

• Model accuracy ~80% (sufficient for real operations).
  
• Recommendation time reduced from minutes to milliseconds (0.03 s / 100 queries).
  
• Manual workload reduced by approx. 30%.

“Data Mind developed an NLP classification model for us to assign cases to the right specialists with high accuracy. The solution has proven to be highly effective and efficient in predicting which specialist an animal should be referred to, based on its condition description, with overall accuracy of about 80% and a fast response time of 0.03 seconds. The model helped us quickly and efficiently assign patients to the right experts, allowing staff to focus more on customer care than administration.”

• Well-prepared and cleaned data.
  
• Focus on speed and cost-efficiency instead of only maximum accuracy.

• Data imbalance between specializations → addressed by balancing the dataset and adjusting metrics.
  
• Real-time requirement → FastText was chosen instead of large language models like BERT.

Companies should consider not only accuracy but also operational costs and real-world usability when choosing an AI model. Having high-quality data and involving the business owner in evaluation is critical.