Oleh O, AI/ML Engineer and Data Scientist

Work Experience

Caller - VoIP server for call center

Implemented voice-to-voice pipeline in Hebrew for VoIP server for call center.

Responsibilities:

• Implemented voice-to-voice pipeline Whisper (STT) - Llama (text-only) - mms-tts (TTS) - OpenVoice (voice cloning),
• Implemented API to interact with models and to integrate them with the VoIP server,
• Prepared dataset for training text-to-speech and speech-to-text models,
• Trained whisper model,
• Deployed models using Docker, created Dockerfiles for models,
• Prompt engineered LLama,
• Configured Asterisk VoIP server,

Tools and technologies: Docker, AWS, ollama, vLLM, Llama, Whisper, MMS-TTS, OpenVoice, Prompt Engineering, Asterisk

Google Doc AI Wrapper

Implemented wrapper for Google Doc AI service in order to improve service accuracy in recognizing freeform documents

Responsibility:

• Used image preprocessing techniques to improve document recognition by Google service,
• Parsed recognized document data and searched for desired form fields, tables and other valuable data,
• Implemented processing and filtering recognized table data,
• Implemented optimization function that combine different image preprocessing methods in order to find combination that provide most recognized data that are searched for

Tools and technologies: Google document ai, OpenCV, pillow, Pandas

Assets tracking with camera POC

tracking asset’s location: real-time positioning, collecting information about cargo and handling equipment using machine vision

Responsibilities:

• Trained container segmentation model, added OCR,
• Created a pipeline for localizing an object on a map using a depth camera,
• Implemented optimization function that combine different image preprocessing methods.

Tools and technologies: Google Vision API, OAK-D (OpenCV Kit), Tensorflow, Keras, Unity3D (synthetic data generation, visualization)

LiveScetchScanner

Implemented camera image and mask processing module; implemented voice command interface

Responsibilities:

• Extraction image part based on mask from NN and Hough line detector,
• Image transformation, processing, filtering,
• implementation voice command interface using Amazon Lex bot API and rhino model

Tools and technologies: OpenCV, Pyaudio, Pvporcupine, rhino, Amazon Lex, boto3

Meowtalk

App to translate the cat’s meows

Responsibilities:

• Created Docker containers with code parts needed for model training,
• set up training pipeline for Meowtalk model using Kubeflow Pipeline in GCP,
• Modified data preprocessing to achieve better accuracy of the model.

Tools and technologies: Docker, Kubeflow pipeline, GCP, Keras

Creating the model to predict the playoff refsults of the World Cup

Responsibilities:

• Data analysis, preprocessing, data cleaning
• Search for additional data
• Feature development based on all available data
• Dataset formation for training models
• Selection of models (Linear models, Decision trees, MLP) for prediction and analysis of results

Tools and technologies used: scikit-learn, PCA, Linear models, Decision trees, MLP

Meeting summarizer

Implemented model for creation summaries of audio meetings based on article

Responsibility and achievements:

• Research of existing abstractive text summarization methods
• Implemented tree-based method

Tools and technologies used: Python, Azure, spacy, NLP-abstractive text summarization, MIP

Deforestation detection

Developing model for deforestation detection on Ukraine territory

Achievements:

• Collected dataset of map fragments: writing scripts to work with the API, searching for the necessary fragments by coordinates, date, filtering by cloudiness and types of fragments and their download
• Created a dataset for network training by clustering images (k-means, image clustering), searching for outliers / anomalies, and then selecting images from different clusters
• Implemented forest watch service based on Azure function
• Implemented a model for deforestation detection.

Tools and technologies used: Rasterio, MS Azure, image clustering

Education

• Master’s Degree in Computer Science, National University