Junior Data Scientist – Deep Learning, Machine Learning & Self-Hosted LLMs

Job Description: Data Scientist – Deep Learning, Machine Learning & Self-Hosted LLMs
Department: Data Science / Artificial Intelligence
Experience Required: 0–1 year of applied Machine Learning experience
Employment Type: Full-time
Job Level: Entry-Level / Junior Data Scientist
About the Role
We are looking for a technically strong and motivated Data Scientist with a foundation in Deep Learning, Machine Learning, Computer Vision, and Large Language Models (LLMs). The ideal candidate should be comfortable developing, evaluating, and deploying machine learning models, working with open-weight LLMs, and building reliable AI solutions for real-world business problems.
This role offers an opportunity to work on practical AI applications, including self-hosted language models, computer vision systems, edge AI deployment, model evaluation, and machine learning engineering. You will work with modern AI frameworks and data tools to build solutions that are reproducible, measurable, and suitable for production or stable proof-of-concept environments.
Roles and Responsibilities
1. Machine Learning and Deep Learning Development
Develop, train, fine-tune, and evaluate supervised machine learning and deep learning models.
Work with PyTorch, Hugging Face Transformers, scikit-learn, and gradient-boosting frameworks.
Prepare datasets, engineer features, conduct exploratory data analysis, and identify data quality issues.
Build reproducible training pipelines with appropriate configuration, experiment tracking, and version control.
2. LLM Fine-Tuning and Self-Hosted AI Solutions
Fine-tune open-weight Large Language Models and embedding models using techniques such as LoRA and QLoRA.
Work with Hugging Face PEFT/TRL and relevant fine-tuning tools.
Support the deployment and evaluation of self-hosted models through internal APIs or inference endpoints.
Develop LLM-powered applications using retrieval-augmented generation (RAG), prompt engineering, and embedding-based search.
Evaluate whether fine-tuning, retrieval, or prompt optimization is the most suitable approach for a given business problem.
3. Model Evaluation, Calibration, and Monitoring
Design evaluation datasets, gold-standard test sets, and appropriate training and validation splits.
Identify data leakage, class imbalance, and other factors that affect model performance.
Evaluate models using metrics such as PR-AUC, precision, recall, Brier score, Expected Calibration Error (ECE), and reliability diagrams, where applicable.
Calibrate model probabilities and select decision thresholds based on agreed business requirements.
Monitor model performance and data drift, document known limitations, and prepare model cards covering training data, evaluation results, risks, and accuracy constraints.
4. Computer Vision and Edge AI
Develop and evaluate computer vision models for object detection, image segmentation, and optical character recognition (OCR).
Work with architectures and tools such as YOLO, RT-DETR, SAM, Vision Transformers (ViT), OpenCV, and torchvision.
Build data annotation, preprocessing, and augmentation pipelines.
Measure model performance using metrics such as mean Average Precision (mAP) and Intersection over Union (IoU).
Support model optimization and deployment on edge devices, evaluating accuracy, inference latency, and resource consumption against server-based performance.
5. Data Engineering and Analytics
Process structured and unstructured data using Python, pandas, Polars, and DuckDB.
Write SQL queries and build feature tables for machine learning applications.
Work with Parquet, Apache Arrow, and dbt for efficient data processing and transformation.
Handle nested documents and complex data structures while maintaining data quality and reproducibility.
6. LLM Application Development and Retrieval Systems
Develop LLM-based applications using frameworks such as LangChain, LangGraph, or LlamaIndex.
Integrate vector databases such as FAISS, pgvector, Qdrant, or Milvus.
Evaluate retrieval and generation quality using tools such as Ragas or DeepEval.
Understand hosted LLM APIs, structured outputs, prompt caching, and batch processing.
7. Model Deployment and MLOps
Write modular, readable, typed, and testable Python code following good software engineering practices.
Develop and test model-serving APIs using FastAPI.
Use Git, Docker, Linux, Bash, and CI/CD workflows to support model development and deployment.
Track experiments using MLflow or Weights & Biases and maintain versioned datasets and model artifacts.
Write unit tests, maintain configurations, and support reproducible model training and deployment.
Experience Required
0–1 year of applied Machine Learning experience, with practical exposure to building, evaluating, or deploying ML models.
Experience with at least one supervised machine learning project in which model predictions or calibrated probabilities supported a human decision.
Academic projects, internships, research work, open-source contributions, or independently developed projects may be considered when they demonstrate relevant hands-on skills.
A strong understanding of model evaluation, data preprocessing, and fundamental machine learning concepts is expected.
Required Technical Skills
Programming and Data Science
Python and SQL.
PyTorch and scikit-learn.
Hugging Face Transformers, Datasets, and Accelerate.
pandas, Polars, DuckDB, and experience with structured and unstructured data.
XGBoost, LightGBM, or CatBoost.
Deep Learning and LLMs
Fundamentals of neural networks, CNNs, and transformer architectures.
Understanding of LLMs, embeddings, RAG, and prompt engineering.
Hands-on exposure to fine-tuning techniques such as LoRA or QLoRA.
Familiarity with Hugging Face PEFT/TRL or equivalent tools.
Computer Vision
Understanding of object detection, segmentation, and OCR.
Familiarity with OpenCV, torchvision, Ultralytics, or similar frameworks.
Knowledge of image preprocessing, augmentation, and model evaluation.
Model Evaluation
Understanding of classification metrics, precision-recall trade-offs, class imbalance, and data leakage.
Familiarity with probability calibration, threshold selection, and model monitoring.
Ability to interpret model performance and communicate limitations clearly.
Engineering and Development
Git and basic software development workflows.
Testing, debugging, type hints, and

Work arrangement
Yes