Prompt Engineer / Agent LLM Specialist

DCG is a space where business needs and people's ambitions meet. They help candidates find environments to develop their skills and assist companies in building effective teams by closely listening and responding to what matters to them, creating lasting and valuable relationships.

• Experience in projects with agent-based LLM systems in regulated environments (e.g., banking) including PoC, pilot, and UAT
• Practical knowledge of Azure OpenAI in enterprise environments and experience with reasoning models (e.g., o-series, Claude Extended Thinking)
• Designing and developing prompts as "prompt as code" — versioning in Git, code review, CI/CD, regression tests, dev/test/prod environments
• Designing modular prompts and agent architectures (ReAct, planning, self-check, multi-agent orchestration)
• Knowledge of RAG and grounding — designing solutions based on knowledge sources, regulations, procedures, and client documents
• Creating prompts resistant to hallucinations, prompt injection, and other security risks
• Integrating AI solutions with business systems and agent orchestration tools
• Evaluation methods for models and prompts: golden datasets, eval sets, regression testing, LLM-as-a-judge, quality and stability metrics
• Experience with tools like LangSmith, Langfuse, Braintrust, Promptfoo or own evaluation frameworks
• Analyzing agent traces, debugging failure modes, and iterative improvement based on runtime data

Nice to have:

• CI/CD
• Azure Open AI
• RAG

• Designing prompts resistant to hallucinations; separation of trusted/untrusted data
• Modular prompts: roles, instructions, constraints, output schema; reusability, parameterization, versioning
• System of prompts for agents with tool access; precise tool descriptions
• Designing outputs compatible with downstream: JSON, decision schemas, tables, validatable structures
• Multi-agent orchestration: supervisor/worker patterns, agent specialization
• Context management in long trajectories: compression, summaries, memory strategies
• Prompts for RAG: citation patterns, grounding on sources, handling "no data"
• Designing retrieval and context compression strategies per use case
• Building golden datasets and evaluation sets for prompts and agent trajectories
• Defining and measuring metrics: response quality, task success rate, tool-call accuracy, trajectory quality, stability, cost, time
• LLM-as-a-judge: usage considering limitations and bias
• Regression testing for prompt/model/API changes; participation in UAT
• Working with Braintrust, LangSmith, Langfuse, Promptfoo or own evaluation harnesses
• Identifying and mitigating risks: prompt injection, data exfiltration via tool use, jailbreaks, data disclosure