Data Scientist (LLM/RAG, project based)

Data Science UA is a service company with deep expertise in AI and Data Science. Our story began in 2016 with the first Data Science UA Conference in Kyiv, and since then we've built one of the largest AI communities in Europe.

About the role:
We are looking for a Data Scientist to design and build data-driven and machine-learning capabilities for clients products - a global leader in Advertising Resource Management. 
You will work with Product, Engineering, and domain experts to translate business questions into analytically sound agentic solutions. The role covers the full data science lifecycle: problem definition, data exploration, experimentation, modelling, evaluation, and production-readiness assessment.

Responsibilities:
- Agentic Solutions & AI Orchestration (Core Focus)
- Autonomous Workflow Design: Architect, build, and deploy autonomous and semi-autonomous AI agents (using modern LLM orchestration, tool-use, and multi-agent frameworks) to automate complex, multi-step media management workflows.
- Tool-Using Agents: Develop agents capable of interacting directly with client's internal tools and external APIs (e.g., DSPs, ad servers, ERPs) to execute actions, run diagnostics, and fetch real-time campaign data.
- Human-in-the-Loop (HITL) Guardrails: Design robust evaluation, permissioning, and safety frameworks to ensure agentic decisions—especially those involving live budget allocations or financial adjustments—operate safely with appropriate human oversight.
- Media Planning & Predictive Intelligence
- Agent-Driven Media Planning: Build predictive frameworks (e.g., MMM, reach & frequency forecasting) and feed them into autonomous planning agents that can generate, critique, and optimize cross-channel media plans automatically.
- Scenario Simulation: Develop models that allow AI agents to simulate "what-if" budget scenarios, channel mixes, and KPI trade-offs to present optimized strategies to media planners.
- Autonomous Buying & In-Flight Optimization
- Real-Time Execution Agents: Create autonomous optimization agents that continuously monitor in-flight campaign performance, automatically reallocating budgets, adjusting pacing, and executing "next-best-action" optimizations.
- Multi-Platform Analytics: Model performance trends across programmatic platforms, direct publisher buys, and walled gardens to feed real-time decision loops for buying agents.
- Agentic Financial Reconciliation & Audit
- Automated Audit Agents: Build autonomous auditing agents that cross-reference planned spend, ad server delivery logs, publisher invoices, and contract terms to detect discrepancies and flag billing leakage.
- Anomaly & Leakage Detection: Design predictive algorithms for automated discrepancy resolution, overspend prevention, and financial clearing integrations (e.g., SAP, ERPs).
- Technical Leadership & Agent Governance
- Agent Observability & Evaluation: Establish team standards for agent benchmarking, prompt/tool evaluation, latency optimization, and failure mode analysis in production environments.
- Mentorship & Strategy: Mentor data scientists on modern LLM architectures, agentic design patterns (RAG, multi-agent systems, tool execution), and production deployment best practices. 

Requirements:
- Agentic AI & Modern LLM Architecture
- Agentic Frameworks: 3+ years of Data Science / ML experience, with hands-on expertise building and deploying production-grade agentic architectures using frameworks like LangGraph, CrewAI, AutoGen, or LlamaIndex.
- Tool Use & Function Calling: Proven experience implementing function calling, tool use, structured outputs (e.g., Pydantic/Instructor), and multi-agent coordination.
- LLM Engineering & RAG: Deep practical understanding of prompt engineering, fine- tuning, retrieval-augmented generation (RAG), context management, and vector database architectures (e.g., Qdrant, Pinecone, Milvus, PGVector).
- AI Guardrails & HITL: Experience designing safety mechanisms, rate/cost controls, human-in-the-loop (HITL) approval gates, and deterministic fallbacks for autonomous agent execution.
- Machine Learning, Statistics & Optimization
- Classical ML & Optimization: Proficiency in supervised/unsupervised ML (Scikit- learn, XGBoost/LightGBM) and mathematical optimization (e.g., SciPy, Pyomo, Google OR-Tools) for resource allocation, pricing, and budget pacing algorithms.
- Time Series & Anomaly Detection: Demonstrated experience building forecasting models (Prophet, ARIMA, NeuralProphet) and anomaly detection frameworks for automated financial/data audit workflows.

Will be a plus:
- Master’s degree or PhD in a quantitative field.
- Experience in advertising, marketing technology, or media optimisation.
- Experience with causal inference, uplift modelling, or marketing measurement.
- Familiarity with AWS, ETL processes, Docker, CI/CD, and model-monitoring tools.