dilara -

• Anomaly Detection – Aerospace & Energy (2022–2023) Developed anomaly detection systems for satellite manufacturing and solar panel maintenance using unsupervised ML techniques (Isolation Forest, Autoencoders, One-Class SVM, LSTM). Improved predictive maintenance and quality control by reducing false positives. • Quality Prediction & Formulation Optimization – Cosmetics & Food (2022–2025) Built end-to-end ML pipelines for predicting product quality from lab data using multi-output regression. Applied MLOps (CI/CD, monitoring, deployment) to deliver a cloud-based API, accelerating formulation testing and time-to-market. • Demand Forecasting & Recommendation – E-commerce (2023, 6 Months) Developed LSTM/RNN-based models for multi-item demand forecasting. Engineered time-series features to improve inventory planning. Contributed to a recommendation system using collaborative filtering and clustering to boost user engagement. • Customer Segmentation & High-Potential Client Prediction – Insurance (2024–2025) Designed ML pipeline for segmenting customers and predicting high-potential leads using structured demographic and transactional data. Applied classification models (Random Forest, XGBoost) with MLOps for model deployment and marketing integration. • Question-Answering System with LLMs – EdTech (2024–2025) Built a knowledge-based QA system using NLP and LLMs (GPT, BERT, Haystack, RAG). Designed a FastAPI-powered chatbot integrated with a custom knowledge graph to enhance user interaction and support. • Proficient in tools like NumPy, Pandas, PyTorch, Statsmodels, Scikit-learn, Keras, TensorFlow, XGBoost, LightGBM, CatBoos, matplotlib, OpenCV, SciPy, and Arima family.

• Research on Sentiment Analysis Techniques: Conducted extensive research on sentiment analysis techniques, including lexicon-based approaches, NLP transformers, and evaluation methods for sentiment analysis to understand and enhance analytical accuracy. • Development of a Novel Sentiment Transfer Method: Developed a unique sentiment transfer method from scratch by integrating lexicon-based methods, sentiment classification, and advanced NLP transformers such as Bert and FinBert, enabling innovative applications in sentiment analysis. • Financial Sentiment Analysis Using FinBert: Utilized the FinBert model to effectively label a financial statement dataset, significantly improving the accuracy and relevance of sentiment analysis within the financial domain. • Advancements in Sentiment Transfer Research: Contributed to the advancement of sentiment transfer research by modifying and optimizing an experimental study, specifically tailored to meet project objectives and dataset requirements. • Analysis and Insight Generation from Experiments: Collected and thoroughly analyzed experimental results, delivering valuable insights and actionable findings to guide research and decision-making processes.

• Training camp organized at Sapienza University. The aim was to build an image search engine method based on Neural Networks (3rd place in the competition).

• In the predictive maintenance project for solar panels, various machine learning and data analysis tools were utilized, including NumPy, Pandas, Statsmodels, Scikit-learn, Keras, and TensorFlow. SQL, specifically PostgreSQL, was employed to manage and analyze large datasets. By applying predictive modeling techniques, maintenance schedules were optimized, leading to increased efficiency and reduced costs.