Mykola Kyrychenko

Lawrence Startup – Legal Industry Platform DevelopmentAt Lawrence Startup, I engineered a sophisticated GenAI platform designed to navigate the complexities of the Swiss legal industry. This platform featured an Agent-Based LLM system utilizing CrewAI and LangGraph, supported by OpenAI's GPT-4 and GPT-4o API. I spearheaded the integration and data extraction from multiple sources, including a custom-built Azure CosmosDB and open web sources, enhanced by Retriever-Augmented Generation (RAG) implemented via custom Azure Functions. Additionally, I was responsible for building the CosmosDB and utilized LangSmith for rigorous platform evaluation.

HappySmile – Smart Psychologist ChatbotAs a GenAI engineer at HappySmile, I developed a smart psychologist chatbot aimed at addressing mental health issues. This project involved orchestrating a sophisticated infrastructure to intelligently manage diverse user interactions. I leveraged the Whisper model to create a robust dataset from extensive video therapy sessions. Furthermore, I fine-tuned several LLMs including OpenAI GPT-3.5, Llama2, Mistral, and Llama3, ensuring the chatbot responded with empathy and effectiveness akin to a human therapist.

Technologies: OpenAI models, OpenAI Embeddings, Langchain, CrewAi/LangGraph, RAG, WeaviateDB, Pinecone, Llama2/Llama3 models, transformers, streamlit, Azure, AWS

I led the development of an advanced internal chatbot designed to facilitate communication with e-learning documents, integrating OpenAI GPT models, Llama2, and RAG. This chatbot supported both English and German, enhancing accessibility and user interaction with educational content. Additionally, I developed an advanced search feature for e-learning documents, which included a visual representation of documents with highlighted sections indicating where answers to queries were located.

In the realm of candidate assessment, I implemented a review and scoring system where a LLM (GPT-4-turbo API) generated personalized questions based on user applications. Candidates interacted with this system through a chatbot interface to provide their answers. The system then ranked candidates using a scoring system that visually represented their qualifications and suitability.

Throughout these projects, I was responsible for the quantization, fine-tuning, and deployment of models to ensure they were optimized for specific e-learning formats. These models were deployed and scaled on Azure and Telekom GPUs, ensuring high performance and scalability to meet the demands of our e-learning initiatives.

Technologies: OpenAI models, OpenAI Embeddings, Langchain, LlamaIndex, RAG, FAISS, Llama2/Llama3 models, transformers, streamlit, Azure, AWS, Telekom, CUDA, Sentence Transformers, BERT, Dolly.

• Connectors Data Engineering
  - Developed custom data collectors to aggregate large-scale datasets from various social media and advertising platforms like Facebook, DV360, Snapchat, and Twitter. Utilized data engineering techniques to manage the influx of data, ensuring efficient preprocessing and transformation using BigQuery and Dataflow.
• Data Platform Migration
  - Played a key role in the migration of data systems to new platforms, ensuring seamless integration and continuity. Collaborated with cross-functional teams to adapt anomaly detection systems and other machine learning solutions to the updated architecture, maintaining robust data integrity and operational efficiency.
• ML Models for Statistics and Anomaly Detection
  - Designed and implemented advanced machine learning models for statistical analysis and anomaly detection. Focused on identifying irregular patterns and potential fraud within e-commerce contexts, using machine learning frameworks in GCP and Databricks for effective model evaluation and refinement. Worked closely with data scientists and domain experts to tailor models to the strategic needs of the business, enhancing real-world applicability and decision-making capabilities.

As part of my role at the startup, I was responsible for the end-to-end flow of our fraud detection system and customer lifetime analysis modeling, which focused on identifying suspicious behavior in Shopify e-commerce shops. My tasks ranged from data engineering to model deployment. I spearheaded efforts in dimension reduction to enhance model efficiency and accuracy. Additionally, I developed the backend for alert triggers, creating a robust system that notifies administrators of potential fraudulent activity in real-time. This comprehensive approach not only improved security but also supported strategic business decisions through detailed customer analytics.

Technologies: AWS Services (RDS, Lambda, Neptune, Sagemaker, AuroraDB/PostgreSQL, Gremlin, S3, XRAY), Boto3, Scikit-learn, Numpy, Pandas, Sagemaker, pyplot, PCA, Docker

As a Machine Learning Developer, I spearheaded the digitalization of complex, unstructured PDF files, employing advanced Computer Vision techniques for object detection to identify text areas and graphical components. I utilized the BERT model to deepen the semantic analysis and contextual interpretation of the text extracted from these documents.

Technologies: BERT Multilingual, YOLOv3, R-CNN, Fast R-CNN, Faster R-CNN, opencv

As a Machine Learning Developer, I spearheaded the digitalization of complex, unstructured PDF files, employing advanced Computer Vision techniques for object detection to identify text areas and graphical components. I utilized the BERT model to deepen the semantic analysis and contextual interpretation of the text extracted from these documents.

Technologies: Python, Flask, PostgreSQL, Redis, Git