Data Scientist

About JiBe:

JiBe is a cloud based fully integrated ERP system for the shipping industry. JiBe ERP enables increased automation and streamlining of processes, creating pre-defined work flows and reducing the usage of email and paper.

Job Responsibilities:

Agentic AI System Development

• Design and develop sophisticated agentic AI systems that orchestrate multi-step reasoning, decision-making, and tool use to solve complex real-world problems
• Integrate diverse model types — including large language models accessed via
• OpenRouter, fine-tuned transformer models, and classical ML models — as tools within agentic workflows
• Continuously evaluate and incorporate emerging agentic frameworks, patterns, and best practices into the team's solutions

Optical Character Recognition (OCR)

• Develop and improve advanced OCR solutions addressing highly complex and varied document types
• Work on document and page classification, determining document types, layouts, and structures as a foundation for downstream processing
• Design and implement sophisticated information extraction pipelines that identify, parse, and structure data from unstructured or semi-structured documents with high accuracy and reliability

Retrieval Augmented Generation (RAG)

• Contribute to the development of RAG solutions that leverage data extracted through the team's OCR and information extraction pipelines
• Collaborate closely with the data engineering team on the design of retrieval pipelines, vector stores, and data preparation workflows that underpin RAG systems

Collaboration & Integration

• Work as an integrated member of a strong, established data science team, contributing expertise while aligning with shared architectural and methodological standards
• Collaborate closely with data engineers to define data requirements, provide feedback on pipeline outputs, and ensure data consumed from Databricks and MongoDB meets the needs of model development
• Participate in code reviews, knowledge sharing, and the continuous elevation of the team's technical standards

Technical Skills

• Deep expertise in Python, including advanced concepts such as async/await concurrency, decorators, context managers, metaprogramming, type hinting, and design patterns. Proven ability to write clean, maintainable, and performant code following best practices
• Strong hands-on experience building scalable, production-grade microservices using FastAPI. Proficiency in API design, dependency injection, middleware integration, async endpoint development, OpenAPI/Swagger documentation, and performance optimization.
• Solid experience with the ML/AI ecosystem: Hugging Face Transformers, scikit-learn, and PyTorch or TensorFlow.
• Familiarity with ML-Flow, model serving, containerization (Docker), and orchestration (Kubernetes) for ML workloads.
• Experience training, fine-tuning, and evaluating transformer-based models as well as classical supervised and unsupervised models.
• Solid understanding of prompt engineering strategies including few-shot learning, chain-of-thought reasoning, system prompts, and prompt templating.
• Experience optimizing prompts for accuracy, latency, and cost efficiency including LLM evaluation, and best practices for integrating LLMs into production systems.
• Hands-on experience with LangChain, LangGraph, and other popular LLM orchestration frameworks (e.g., LlamaIndex, Haystack) for building agentic workflows, RAG pipelines, and complex multi-step LLM applications.
• Familiarity with OpenRouter or equivalent LLM gateway/routing platforms is a serious advantage.
• Experience with vector databases (e.g., Milvus, Qdrant, Chroma) for semantic search, retrieval-augmented generation (RAG), and efficient similarity search at scale.
• Experience with Retrieval Augmented Generation (RAG) — including chunking strategies, embedding models, vector search, and retrieval evaluation — is a serious advantage
• Familiarity with MongoDB or any other NoSQL database is an advantage
• Experience working with Databricks or similar large-scale data platforms is an advantage