Stanislav – Python, PyTorch, Scikit-learn

The product aims to help journalists write articles, especially based on other materials, by utilizing generative AI capabilities. Another feature is connected with various automation tools for a company to repair articles: linking to rules from reference books, searching, answering user questions, etc.

• Integrated LLM-based solutions and vector search to enhance content retrieval and search efficiency across large datasets
• Designed and implemented pipelines using Temporal and Kafka for persistent, scheduled data processing, leveraging Regex and LLM evaluation for accurate data extraction from external APIs
• Developed ingestion pipelines to preprocess and index external data into Typesense and MongoDB, optimizing it for advanced search and retrieval tasks
• Build the RAG system that integrates 3 data sources to provide users with answers based on the stored data
• Collaborated on building APIs to support seamless integration of machine learning capabilities with client applications, ensuring robust server-client communication

Multichannel AI network with services for compute providers, AI model users, and enterprise clients. The platform provides access to the robust models for the users, or connect to the network for the providers to start earning ERC 20 tokens.

• Designed and implemented the natural language processing (NLP) pipeline via NLTK and spaCy to process and analyze large volumes of text data for sentiment analysis, entity recognition, and topic modeling
• Developed ML models for text classification and clustering using Scikit-learn and TensorFlow
• Implemented the interaction layer to enable seamless communication between the provider, AI inference flow, and the user
• Implemented WebSocket exchange for AI inference, status handling, and uploading task results into S3
• Implemented system metrics monitoring for API infrastructure, user & provider analytics, and data aggregations

A platform for performing audio and video recording transcription, transcription editing, and analysis. It aims at personal and corporate users. The most significant users are the Norway State Library and the Norway Police.

• Fine-tuned BERT neural network for NLP tasks to support speech-to-text features
• Developer a public API with Python and Flask to provide transcription functionality to corporations, driving user acquisition and product engagement
• Optimized speech-to-text pipeline, making it take half the length of the audio clip
• Designed and integrated a high-performance search functionality for transcriptions by leveraging ElasticSearch for fast querying
• Streamlined the deployment pipeline using CI/CD automation, reducing release times by 90%