ML-powered microservice for customer request processing

Our partner needed a solution to automate incoming customer requests for their sales team. We had to build a microservice-based solution that could extract unstructured data, convert it into a readable format, and automatically fill predefined templates for further processing.

We began building an ML microservice model with the Transformers library and language models, and later used the PyTorch framework for training and improvements. We chose TorchServe to serve the model with API services and deploy it, and used Docker to containerize the microservice to make sure it would be compatible with any platform. One of this project’s major challenges was making certain the model gave precise and contextually accurate responses. To ensure exactness, we refined all extracted data with post-processing.

Final touches were making sure the list of questions could expand to improve the model’s adaptability, helping the microservice evolve and grow as our partner’s business does. In the end, we delivered a robust ML-based service that improves our partner’s customer communication.

Project outcomes

1. Reduced our partner’s load and improved efficiency with a machine learning microservice.
2. Made the microservice scalable to meet growth demands with Docker.
3. Improved end customer satisfaction with a tool for accurate and consistent communication.