Sergii G.

Main Goal: Develop an LLM-powered assistant that can summarize, analyze, and extract key insights from legal documents.

Role: ML Engineer.

Tasks:

• Implement RAG (Retrieval-Augmented Generation) for relevant document retrieval.
• Optimize prompts for legal question-answering.
• Integrate the system with a chatbot or web UI for user interaction.

Technologies: Python, LangChain, OpenAI/Claude, Pinecone/FAISS, FastAPI, Streamlit, Celery, Docker.

Role: ML Engineer

• Created an interactive AI tutor app using the Dify framework, powered by a fine-tuned LLaMA‑3.1 model for personalized education in math and science.
• Configured Dify's RAG modules with FAISS to connect to a structured knowledge base, ensuring answers are grounded in curriculum-aligned material.
• Used Dify’s frontend builder and memory system to track student progress, learning patterns, and recurring questions, enabling personalized follow-up.
• Implemented dynamic prompts and function calling via Dify tools to adapt question difficulty and simulate step-by-step human-like explanations.
• Launched a self-service educational chatbot via Dify’s hosted UI and API interface, accessible both via chat and embeddable widget in learning platforms.

Technologies:  Python, Hugging Face Transformers (LLaMA‑3.1), Dify (App, RAG, Memory, Prompt Management), LangChain, FAISS, NumPy, Pandas, Streamlit, Docker

Role: ML Engineer

• Developed an AI-powered support assistant using the Dify framework to automatically classify, summarize, and route incoming support tickets.
• Fine-tuned a BERT-based model for intent classification and priority labeling, then integrated it into a Dify app for seamless user-facing chatbot interaction.
• Leveraged Dify’s retrieval-augmented generation (RAG) pipeline with SentenceTransformers + Elasticsearch to surface relevant past tickets and their resolutions.
• Used Dify’s workflow tools to integrate agent feedback into the retraining loop, enabling continuous model improvement and adaptive system behavior.
• Deployed the complete pipeline using Dify's FastAPI backend and Docker, making it easy to scale and integrate with internal CRM tools.

Technologies: Python, Hugging Face Transformers (BERT), Dify (App, Dataset, Feedback), SentenceTransformers, Elasticsearch, OpenAI GPT-4o, FastAPI, Docker, Pandas

Description: Developed a machine learning pipeline to predict forest cover types based on cartographic variables using ensemble models. 

Tasks:

• Performed exploratory data analysis to understand feature distributions and correlations, identifying patterns and potential data quality issues.
• Engineered and selected relevant features, using techniques like PCA and statistical methods to reduce dimensionality and improve model efficiency.
• Trained and evaluated ensemble models (Random Forest, XGBoost) with optimized hyperparameters, using cross-validation to ensure robust performance.
• Implemented overfitting mitigation strategies, including regularization and pruning, to enhance model generalization on unseen data.

 Technologies: Python, NumPy, XGBoost, Matplotlib, Seaborn.

Role: ML Engineer

• Built a creative writing assistant that helps authors brainstorm and refine stories using an LLM fine-tuned for long-form narrative consistency and style.
• Added a memory module to track characters, plot points, and tone guidelines, ensuring coherence and continuity across chapters of a novel or screenplay.
• Enabled the assistant to suggest plot twists or character developments through controlled text generation techniques and user-provided cues.
• Developed an interactive web interface where writers can iteratively edit the AI’s suggestions, preserving their own voice while benefiting from AI-driven creativity.

Technologies: Python, Hugging Face Transformers (LLaMa-3.1 fine-tuned), PyTorch, LangChain (memory management), FAISS, NumPy, Gradio, Docker

A multimodal model was trained to predict the adoption rate of animals from shelters. The model use photos of pets and descriptions from the table

Tech stack : Python, Jupyter notebook, numpy, pandas, scikit-learn , cv2, matplotlib, PyTorch, eficientnet_b3 , RobertaModel

Trained model for customer churn prediction

Tech stack: Python, Jupyter notebook, numpy, pandas, scikit-learn

Main Goal: Create an AI-powered tutor that adapts to a student’s learning style, explains concepts, and provides quizzes.

Role: ML Engineer.

Tasks:

• Design a conversational AI model for personalized learning.
• Implement a knowledge retrieval system using vector databases.
• Create dynamic quizzes and feedback loops based on student responses.
• Connect the system to educational resources via APIs.

Technologies: Python, LangChain, OpenAI, FAISS, Streamlit, FastAPI, Hugging Face Transformers.

Description: Built a binary classification model to detect fraudulent financial transactions using logistic regression, supported by robust statistical evaluation and cross-validation techniques.

Tasks:

• Developed and trained a logistic regression model to identify fraudulent activity.
• Applied cross-validation to validate model performance and prevent overfitting.
• Selected appropriate evaluation metrics (e.g., precision, recall, F1-score) for imbalanced data.
• Conducted statistical testing to assess the significance of model features and predictions.

 Technologies: Python, Scikit-learn, Pandas, NumPy, Matplotlib, Seaborn.

Description: Develop a regression model to predict house prices based on various factors such as location, size, and amenities.

Role in the project: ML Engineer.

Responsibilities: Data collection, exploratory data analysis (EDA), feature selection, model training, and deployment.

Technical summary: Python, scikit-learn, pandas, XGBoost, Flask/FastAPI for deployment.

Description: Build a classification model to predict loan default risk using financial and demographic data.

Role in the project: Data Scientist.

Responsibilities: Data cleaning, feature engineering, imbalanced data handling, model selection, and explainability.

Technical summary: Python, scikit-learn, LightGBM, SHAP for interpretability, SQL for data extraction.

Web-based chatbot that can upload PDF documents and provide question-answering capabilities based on the content of uploaded document

Tech stack : Python, Django, LLM ,Pandas, Docker, Alembic, GitHub

Description: Predict whether a customer is likely to stop using a product or service based on historical data.

Role in the project: Data Scientist.

Responsibilities: Data preprocessing, feature engineering, model training, and evaluation.

Technologies: Python, pandas, scikit-learn, XGBoost, TensorFlow/PyTorch, Streamlit for visualization.

Description: Develop a recommendation system for an e-commerce platform to suggest relevant products to users.

Role in the project: Data Scientist.

Responsibilities: Data preprocessing, collaborative filtering/content-based recommendations, model optimization, evaluation.

Technical summary: Python, pandas, scikit-learn, Surprise library, TensorFlow, FastAPI/Flask for API deployment.

Description: Identify fraudulent transactions in financial datasets using machine learning techniques.

Role in the project: ML Engineer.

Responsibilities: Data preprocessing, anomaly detection, building supervised/unsupervised models, real-time deployment.

Technical summary: Python, scikit-learn, TensorFlow/PyTorch, Isolation Forest, Autoencoders, Kafka for streaming.

Main Goal: Create a system that generates stylized images of people, pets, or objects in various artistic styles (anime, steampunk, cyberpunk, etc.).

Role: ML Engineer.

Tasks:

• Fine-tune a Stable Diffusion model using DreamBooth for custom image generation.
• Implement a UI where users can upload images and select styles.
• Optimize inference for faster generation using model quantization or LoRA.
• Deploy as a web app with real-time image processing.

Technologies: Python, Stable Diffusion, DreamBooth, OpenCV, Streamlit, FastAPI, Hugging Face Diffusers, Docker.

Main Goal: Develop a Computer Vision system that detects cars in real time and classifies their make, model, and color.

Role: ML Engineer.

Tasks:

• Collect and preprocess car images for training.
• Train an object detection model (YOLOv8, Faster R-CNN) for car detection.
• Implement a classification model for make/model/color recognition.
• Deploy the system as an API for integration with traffic monitoring or security cameras.

Technologies: Python, PyTorch/TensorFlow, OpenCV, YOLOv8/Faster R-CNN, FastAPI, Docker.

This project was a test task. There is a sphere which moves thru a 3D surface on a specific path and deletes each point it touches. My task was to find this points and save the 1st layer of the surface.

Skills achieved: Basics of OOP Linear algebra

Teaching python for teenagers;

Cooperation with team Students wrote and successfully submitted Telegram bot projects, Flask app projects.

Communication with clients
Non food FMCG sales
Sales analytic
Cooperation with team and other departments
Сontributed to a 10% increase in the turnover of the company's branch