Senior/Middle Data Scientist (Data Preparation & Pre-training)
Data Science UA is a service company with strong data science and AI expertise. Our journey began in 2016 with uniting top AI talents and organizing the first Data Science tech conference in Kyiv. Over the past 9 years, we have diligently fostered one of the largest Data Science & AI communities in Europe.
About the client:
Our client is an IT company that develops technological solutions and products to help companies reach their full potential and meet the needs of their users. The team comprises over 600 specialists in IT and Digital, with solid expertise in various technology stacks necessary for creating complex solutions.
About the role:
We are looking for an experienced Senior/Middle Data Scientist with a passion for Large Language Models (LLMs) and cutting-edge AI research. In this role, you will focus on designing and prototyping data preparation pipelines, collaborating closely with data engineers to transform your prototypes into scalable production pipelines, and actively developing model training pipelines with other talented data scientists. Your work will directly shape the quality and capabilities of the models by ensuring we feed them the highest-quality, most relevant data possible.
Requirements:
Education & Experience:
– 3+ years of experience in Data Science or Machine Learning, preferably with a focus on NLP.
– Proven experience in data preprocessing, cleaning, and feature engineering for large-scale datasets of unstructured data (text, code, documents, etc.).
– Advanced degree (Master’s or PhD) in Computer Science, Computational Linguistics, Machine Learning, or a related field is highly preferred.
NLP Expertise:
– Good knowledge of natural language processing techniques and algorithms.
– Hands-on experience with modern NLP approaches, including embedding models, semantic search, text classification, sequence tagging (NER), transformers/LLMs, RAGs.
– Familiarity with LLM training and fine-tuning techniques.
ML & Programming Skills:
– Proficiency in Python and common data science and NLP libraries (pandas, NumPy, scikit-learn, spaCy, NLTK, langdetect, fasttext).
– Strong experience with deep learning frameworks such as PyTorch or TensorFlow for building NLP models.
– Ability to write efficient, clean code and debug complex model issues.
Data & Analytics:
– Solid understanding of data analytics and statistics.
– Experience in experimental design, A/B testing, and statistical hypothesis testing to evaluate model performance.
– Comfortable working with large datasets, writing complex SQL queries, and using data visualization to inform decisions.
Deployment & Tools:
– Experience deploying machine learning models in production (e.g., using REST APIs or batch pipelines) and integrating with real-world applications.
– Familiarity with MLOps concepts and tools (version control for models/data, CI/CD for ML).
– Experience with cloud platforms (AWS, GCP, or Azure) and big data technologies (Spark, Hadoop, Ray, Dask) for scaling data processing or model training.
Communication & Personality:
– Experience working in a collaborative, cross-functional environment.
– Strong communication skills to convey complex ML results to non-technical stakeholders and to document methodologies clearly.
– Ability to rapidly prototype and iterate on ideas
Nice to have:
Advanced NLP/ML Techniques:
– Familiarity with evaluation metrics for language models (perplexity, BLEU, ROUGE, etc.) and with techniques for model optimization (quantization, knowledge distillation) to improve efficiency.
– Understanding of FineWeb2 or similar processing pipelines approach.
Research & Community:
– Publications in NLP/ML conferences or contributions to open-source NLP projects.
– Active participation in the AI community or demonstrated continuous learning (e.g., Kaggle competitions, research collaborations) indicating a passion for staying at the forefront of the field.
Domain & Language Knowledge:
– Familiarity with the Ukrainian language and context.
– Understanding of cultural and linguistic nuances that could inform model training and evaluation in a Ukrainian context.
– Knowledge of Ukrainian text sources and data sets, or experience with multilingual data processing, can be an advantage given the project’s focus.
MLOps & Infrastructure:
– Hands-on experience with containerization (Docker) and orchestration (Kubernetes) for ML, as well as ML workflow tools (MLflow, Airflow).
– Experience in working alongside MLOps engineers to streamline the deployment and monitoring of NLP models.
Problem-Solving:
– Innovative mindset with the ability to approach open-ended AI problems creatively.
– Comfort in a fast-paced R&D environment where you can adapt to new challenges, propose solutions, and drive them to implementation.
Responsibilities:
– Design, prototype, and validate data preparation and transformation steps for LLM training datasets, including cleaning and normalization of text, filtering of toxic content, de-duplication, de-noising, detection and deletion of personal data, etc.
– Formation of specific SFT/RLHF datasets from existing data, including data augmentation/labeling with LLM as teacher.
– Analyze large-scale raw text, code, and multimodal data sources for quality, coverage, and relevance.
– Develop heuristics, filtering rules, and cleaning techniques to maximize training data effectiveness.
– Collaborate with data engineers to hand over prototypes for automation and scaling.
– Research and develop best practices and novel techniques in LLM training pipelines.
– Monitor and evaluate data quality impact on model performance through experiments and benchmarks.
– Research and implement best practices in large-scale dataset creation for AI/ML models.
– Document methodologies and share insights with internal teams.
The company offers:
– Competitive salary.
– Equity options in a fast-growing AI company.
– Remote-friendly work culture.
– Opportunity to shape a product at the intersection of AI and human productivity.
– Work with a passionate, senior team building cutting-edge tech for real-world business use.
About
