● Developed an advanced chatbot assistant leveraging Retrieval-Augmented Generation (RAG) and Large Language Models (LLM), integrating Qdrant vector database for efficient information retrieval and response generation. Engineered a RAG-based application using Streamlit, incorporating: ● GPT-4 and Claude models for natural language processing. ● Vision models (Haiku, Sonnet, and Gemini) for visual data extraction. ● Optimized data chunking methods to enhance retrieval accuracy and processing efficiency. Spearheaded optimization of the photo moderation and scanning workflow using advanced models: ● Implemented YOLOv7 for detecting individuals and children in user-uploaded photos. ● Utilized EfficientNet and ViT (Vision Transformer) for watermark detection and classification. ● Developed models for classifying and auto-cropping screenshot / collage images. ● Conducted sentiment analysis model training using BERT with various attention mechanisms. ● Assisted in fine-tuning Large Language Models (LLM) using RAGs for improved performance and domain-specific applications. ● A system has been developed to help search for information within a specific document and answer questions based solely on its content. In this case, it was a PDF article. ● To achieve this, I used LlamaIndex. It loads the document, splits it into sections, creates an "intelligent" index, and helps find the most important text fragments based on a query. ● These retrieved fragments are then passed to the Mistral-7B language model, which formulates a clear and accurate answer using only the information from the document.

● Led the development of real-time person detection and matting in a collaborative team environment. ● Engineered a car detection model for a prominent carwash organization. ● Implemented an innovative image clustering model for processing user laptop screenshots, employing diverse deep learning feature extraction methods. Conducted analysis and categorization using clustering models to compute users’ key performance indicators (KPIs) based on clustered data. ● Collaborated on the Layout-LM model for Named Entity Recognition (NER) tasks, contributing to the development of an application. Wrapped the model in ONNX format and integrated it with Azure cloud services. ● Played a key role in building a medical tumor image detection model. ● Applied Optical Character Recognition (OCR) for data extraction from images, performed text analysis, and constructed embeddings for text preprocessing and classification.

● Developed predictive models for churn, income, loan, and client segmentation, contributing to enhanced production outcomes. ● Elevated transaction and money change conversion rates for the bank. ● Improved the anti-fraud model’s F1-score metric from 79 percent to 86 percent, fortifying the bank’s security measures. ● Successfully resolved issues with the bank’s database store and production tables, ensuring seamless operations.