Vasil Yordanov

The project I was involved in focused on developing an LLM application designed to provide a chatbot interface for REST APIs. As the machine learning architect and developer, I played a crucial role in its implementation. I utilized various tools and frameworks to enhance the application's functionality, including Chainlit, LangChain, PaLM, and ChatGPT.

The project I was involved in focused on developing an enterprise document Q&A application powered by LLM technology. As the machine learning architect and developer, I played a crucial role in its implementation. We utilized various tools and frameworks to enhance the application's functionality, including Streamlit, LangChain, PaLM, ChatGPT, and Hugging Face.

Our application aimed to address the challenge of efficiently extracting information from vast amounts of textual data within an enterprise setting. By leveraging LLM technology, we were able to develop a robust question-and-answer system capable of understanding and responding to queries based on the contents of documents.

The project, which aimed to develop an enterprise machine learning operations (MLOps) architecture for a UK-based, multinational technology company, had a duration of three months. As the machine learning architect, I led a team of machine learning and DevOps engineers in the development of the solution. The architecture heavily relied on 3rd-party tools, including Dataiku and Seldon, enabling seamless collaboration and an end-to-end MLOps lifecycle. The developed solution was deployed on Microsoft Azure, allowing for efficient management of the entire MLOps lifecycle, including data management, model training, and deployment.

The social sentiment analysis system was a remarkable project that I had the pleasure of leading as a machine learning architect. Our development team of ML, MLOps, and front-end engineers built a sophisticated solution based on a BERT model implemented in TensorFlow. The system tracks user sentiment on specified topics in near real time and utilizes only server-less components in Google Cloud. The technology stack includes Vertex Pipelines, Cloud Function, Cloud Run, BigQuery, and Cloud Storage.

The project was completed in three months, and it also included the implementation of a management API in Flask.

During this six-month project, our team designed and implemented various landing zones for development, staging, monitoring, and production. As the machine learning architect, I led the development team comprising machine learning and DevOps engineers to develop an enterprise machine learning operations architecture for an Asian-based multinational technology company. Our solution integrated various cloud-native and third-party systems such as Vertex AI, BigQuery, and Cloud Build to enable seamless collaboration and an end-to-end MLOps lifecycle. We also integrated popular machine learning frameworks such as TensorFlow and PyTorch for model training and deployment. Our architecture enabled efficient model management and reproducibility across all stages of development.

In this project, I worked on a serverless system designed to orchestrate Python-source distributions and Jupyter Notebooks on the Google Cloud Platform. Key features include scheduling Python packages and notebooks as cron jobs, ad-hoc execution of Python packages and notebooks, developing a Python API for Python package and notebook management, and integrating with third-party APIs. As a machine cloud architect, I led the development team consisting of machine learning and back-end engineers during the project implementation.

For 6 months, I worked as a machine learning engineer in a team that developed an object detection model using TensorFlow. The model aimed to identify and detect robots on CCTV camera streams, and I was responsible for the error analysis of different models, the design of datasets to enhance model performance, and the setting up of a cloud architecture for distributed training.

As a data and machine learning engineer in a European startup, I worked on a cloud machine learning system for ingesting, processing, and inferring bio-metric and motion data in near real time. Key features included inference of streaming data, batch processing, and cryptographic key management. The system utilized Kafka and Kafka Streams for data ingestion and processing, PostgreSQL for data storage, and APIs for integrating the system with external services. My responsibilities also included training deep learning models in TensorFlow and implementing data and inference pipelines for production deployment.