AI Engineer – #35029

As an AI Engineer, you will be responsible for building production-grade RAG pipelines across both structured and unstructured data, and designing and optimizing agentic workflows, among other tasks.

Requirements

• Building production-grade RAG pipelines
• Indexing and retrieving across both unstructured files (Word, PDF, email, etc.) and structured relational database entries
• Familiarity with indexing techniques such as OCR (multi-modal), Element-Extraction (Text, Table, Charts etc.), Summary Generation, HyQE, Keyword Extraction, Embeddings (Dense, Sparse, Late-Interaction), Named Entity Recognition etc.
• Familiarity with advanced retrieval techniques such as Multi-Stage Retrieval, Content-Security-Policy, Filter-Extraction, Query-Rewriting, HyDE/HyQE, Query-Expansion, Hybrid-Search and Reranking (bi- und crossencoder) etc.
• Generative AI systems and underlying model behavior
• Deep understanding of how modern LLM-based systems work beyond simple API usage
• Knowledge of hyperparameters and model controls such as Temperature, Top-P, reasoning effort, structured output, etc.
• Solid prompt engineering skills, including instruction design, prompt structuring (e.g. XML tags / Markdown), ordering of instructions, separation between system / user prompts etc.
• Prompt management, evaluation, and observability
• Prompt versioning, variants, testing, and iteration using tools such as Agenta or similar
• Common evaluation and retrieval quality metrics such as Recall, Accuracy, F1, MRR, etc.
• Use of observability / tracing tools such as Langfuse via OpenTelemetry (OTEL) or comparable stacks
• Agentic workflows and tool calling
• Understand tool calling in depth and know how to properly design, scope, and separate tools to maximize reliability, maintainability, and overall system value
• Design systems that do not simply expose “everything to the model”, but instead apply tool prioritization, preselection, and contextual narrowing (e.g. reducing a large toolset to only the most relevant candidates for a given task)
• Understand concepts such as memory, state handling, context persistence, and workflow continuity, and know when and how these should be incorporated into agentic systems
• Familiar with agentic architectures and know how to break down complex tasks into smaller, independently executable subtasks
• Understand when workflows should be fully automated versus when Human-in-the-Loop (HITL) patterns are required, and are able to design such processes based on business logic, risk, and practical constraints

Originally posted on Himalayas