JOBILY AI SUMMARY
What you'll do
- Build model evaluation and benchmarking frameworks for LLM, SLM, and agentic AI use cases
- Recommend model selection based on accuracy, latency, cost, safety, explainability, and operational constraints
- Conduct fine-tuning, LoRA / PEFT, prompt optimization, and retrieval optimization experiments for priority business use cases
- Perform distillation and model compression proof points for smaller or more efficient models
- Develop evaluation datasets, test harnesses, golden-answer sets, and regression testing routines for AI applications
- Create observability dashboards and quality feedback loops covering model performance, hallucination risk, drift, cost, and user feedback
- Design target-state patterns for open-weight or self-hosted LLM adoption including architecture, governance, and operational readiness
Key requirements
- Python programming skills and ML/LLM engineering experience
- Expertise in model evaluation, benchmarking, model selection, and regression testing
- Proficiency in fine-tuning, LoRA / PEFT, prompt optimization, and retrieval optimization techniques
- Knowledge of model compression, distillation, and related optimization methods
- Experience with observability tools such as MLflow, W&B, LangSmith, or Langfuse
- Familiarity with LLM, SLM, human evaluation workflows, and related AI model governance
- Computer Science / Engineering / Artificial Intelligence / Data Science / Machine Learningpreferred
Benefits
- Competitive salary and benefits including life/health insurance, performance bonuses, flexible working, share purchase plan, parental leave, continuous training, career coaching, diverse and inclusive culture, and corporate citizenship initiatives.
About the job
At Accenture Newra AI Hub, we are not just building technology, we are redefining how banking operates. As part of our strategic collaboration with Piraeus, Newra is designed to responsibly embed AI at the core of its business, moving beyond experimentation to real-world impact at scale. Built to…


