- Webscraping: Scraping large volume of structured and/or unstructured textual data from web resources.- ETL: Extracting useful information from large volume of texts with regext, NLP and ML techniques, transforming textual data and feeding into consumer pipelines.- Semantic search: Training and/or fine-tuning NLP models for semantic search, and setting up vector search and/or RAG pipelines.- Prompt engineering: Optimizing prompts for to accomplish complex tasks with LLMs and… Show more - Webscraping: Scraping large volume of structured and/or unstructured textual data from web resources.- ETL: Extracting useful information from large volume of texts with regext, NLP and ML techniques, transforming textual data and feeding into consumer pipelines.- Semantic search: Training and/or fine-tuning NLP models for semantic search, and setting up vector search and/or RAG pipelines.- Prompt engineering: Optimizing prompts for to accomplish complex tasks with LLMs and designing LLM flows that fit into larger systems.- LLM apps: Developing complex LLM apps with Langchain.- Open-source LLMs: Implementing optimized deployments for open-source LLMs like LLaMA, Mixtral, Gemma etc. and fine-tuning them for custom tasks and datasets. Show less

-Set up and managed ETL pipelines for raw VPN data, and conducted experiments for model development and evaluation.-Extracted features from VPN logs and developed an deep Autoencoder model for detecting anomalies in user network traffic.-Combined various scaling and normalization techniques and an appropriately chosen loss function to be able to encode and decode different types of features, including one-hot encoded ones. Deployed rule-based statistical models in… Show more -Set up and managed ETL pipelines for raw VPN data, and conducted experiments for model development and evaluation.-Extracted features from VPN logs and developed an deep Autoencoder model for detecting anomalies in user network traffic.-Combined various scaling and normalization techniques and an appropriately chosen loss function to be able to encode and decode different types of features, including one-hot encoded ones. Deployed rule-based statistical models in production.-Designed and developed a machine learning model that enables the detection of phishing emails.-Extracted 90+ features by analyzing URLs and attachments embedded in emails.-Predicted phishing emails with an F1 score of 0.91 Show less

-Conducted research on object detection and image classification models.-Implemented YOLOv5 model for damage detection through images taken from different angles of rented cars (left‑side,right‑side, front,back)