Isac D.

• Developed a fraud detection system for client documents at Hospital Unimed using Python and Vertex AI, enabling automated classification of personal records and enhancing accuracy in fraud prevention.

• Designed and delivered a Proof of Concept (PoC) for an AI-powered assistant to support psychologists during therapy sessions.

• Built pipelines to process and transcribe audio using Whisper and Pyannote, including speaker diarization for precise session analysis.

• Applied LLMs with Map-Reduce and RAG techniques to extract insights, detect emotions, and identify Cognitive Behavioral Therapy (CBT) elements from therapy transcripts.

• Implemented advanced audio denoising and source separation (DSS) techniques to significantly improve transcription quality by removing background noise.

• Generated structured reports and comprehensive summaries by combining LLM-driven summarization with map-reduce frameworks, effectively addressing context length limitations in large models.

• Developed interactive dashboards using Python, Plotly, Seaborn, and Dash to visualize insights and statistics from therapy sessions, including the recurrence of emotions, frequent cognitive distortions, and other key behavioral metrics.

Developed a Data Visualization tool using Python and Streamlit to analyze magnetic signals obtained from several types of magnetometers (National Instruments, Oros, Meda, Narda) in order to detect electromagnetic interference (EMI).

• Created interactive graphs depicting amplitude versus time, filtered time, and amplitude versus frequency (FFT) using Plotly, facilitating in-depth signal analysis.
• Engineered AC/DC digital filters to reduce noise, optimizing the accuracy of EMI detection using Scipy.
• Implemented a decimation process to effectively manage large EM signals.
• Performed signal processing analysis using Pandas and Numpy.

• Collaborated with technical team using GitHub to improve a defect detection system designed for manufacturing customers.
• Implemented microservices using FastAPI, Flask, and gRPC to process large (10k x 8k pixel) images and apply them into deep learning models.
• Created Python package that utilized a third-party API to streamline the annotation process.
• Implemented unit and integration tests using Docker and Python to improve the quality of delivered code.

• The research aimed to combine tensor techniques with time series forecasting for route prediction of suspect vehicles using sensor data.

The goal of the project was to develop a system for automating processes at a Big Tech from US using machine learning techniques. Specifically, the system was designed to help managers to give access to internal applications and optimise the response time for it.

• Created a recommendation engine using machine learning models with a rejection option over highly imbalanced datasets. Tasks included data visualization, Python programming, data cleaning/processing, feature engineering and selection, model training and evaluation, data analysis, and data ETL using Python;
• Performed feature engineering on highly imbalanced datasets from various data sources such as AWS S3, PostgreSQL, MySQL, and Cassandra;
• Handled the full data science cycle, from feature engineering to model deployment;
• Built a recommendation system to assist upper management with virtual asset access control decision-making;
• Created, evaluated, deployed, and maintained machine learning models as web services;
• Implemented techniques to optimize models, including feature engineering and selection, redundancy detection, outlier detection, over- and under-sampling, model calibration, and dataset drift detection;
• Designed data pipelines using Python to process financial data and migrate data between systems.