We are growing the development team behind one of our client's most successful products — an integrated delivery platform built on the depth of real-world engagement experience. It enables cross-team collaboration, real-time transparency, and better decision-making.

Your job is the AI layer of that product: end-to-end LLM pipelines, retrieval, and multi-agent orchestration running in production, not in a demo. Expect deep Python work, real evaluation metrics, and full ownership of how the system behaves under load, cost, and latency constraints.

• LLM pipeline architecture: design end-to-end pipelines using RAG, embeddings, and orchestration layers
• Multi-agent systems: build modular, scalable agent workflows in LangChain, LangGraph, CrewAI, or AutoGen, including observer and fallback agents
• Retrieval engineering: build chunking and ingestion pipelines for PDFs and unstructured data, and deploy vector stores with semantic search and reranking
• Model strategy: evaluate model options against performance, cost, and capability trade-offs, and define embedding strategies, context windows, and prompt structures
• Evaluation & QA: define and track metrics for GenAI and RAG systems — faithfulness, precision, recall, semantic similarity — using RAGAS, sklearn, and NumPy
• Observability: instrument the system with LangFuse, Datadog, or custom tracing to monitor, debug, and optimize
• Delivery engineering: maintain CI/CD pipelines, Docker containers, test coverage, and API integrations
• Communication: turn complex models into clear insights for executive stakeholders, lead design sessions, and mentor other developers

• 8+ years in software engineering, with at least 3 years focused on AI/ML, NLP, or LLM-based applications
• Expert-level Python — modular, class-based, production-grade code you can debug across distributed services
• Proven GenAI delivery: prompt engineering, embedding-based retrieval, and multi-agent orchestration patterns in production
• Modern AI stack: LangChain, LangGraph, CrewAI, AutoGen, Hugging Face
• Vector databases: Azure AI Search, FAISS, or Postgres with pgvector
• Engineering foundations: Jupyter, Docker, GitHub, REST API integration, CI/CD via GitHub Actions
• AI-assisted development: comfortable with Cursor, GitHub Copilot, and similar tools without sacrificing security or maintainability
• Preferred: multimodal LLMs, ReAct-style prompting, planner-executor agents, human-in-the-loop RAG evaluation

• A production product with real users — no proof-of-concept purgatory
• Full ownership of the AI architecture, from retrieval strategy to observability
• A 12-month engagement with a strong likelihood of extension
• Fully remote work with a team that meets in person roughly once a quarter in Prague
• Direct exposure to executive stakeholders and the reasoning behind product decisions

• Start: ASAP
• Workload: Full-time, 12-month contract with a strong likelihood of extension
• Setup: fully remote, occasional in-person team sessions in Prague (approx. 1× per quarter)
• Overlap: minimum 14:00–18:00 CET with the US team
• Selection process: includes a HackerRank challenge

‍