ML Systems Engineer — Benchmark-Driven Agentic AI

IC Resources; City of Edinburgh; Hybrid; GBP 65,000 - 100,000; Full time

Job summary

IC Resources is recruiting a Machine Learning Engineer for an Edinburgh-based AI startup. The role blends software modelling with benchmarking to validate system behaviour in a hybrid, UK-wide context.

You will develop agentic AI workloads, build a robust benchmark suite, and design experiments to measure how the semantic memory layer affects inference performance. Strong Python and distributed-systems know-how are essential.

Qualifications

• Commercial experience in AI systems, retrieval, or AI infrastructure with shipped production software.
• Hands-on with agentic pipelines, LLM fine-tuning, RAG/GraphRAG, or knowledge graphs.
• Strong Python and comfort across ML, distributed systems, and performance engineering.
• Track record building benchmarks/eval frameworks with rigour.

Responsibilities

• Own the software model used to evaluate system behaviour ahead of hardware.
• Build agentic AI and GraphRAG workloads showing measurable improvements.
• Develop and maintain a benchmark suite (latency, GPU utilisation, token reduction).
• Design experiments to isolate the semantic memory layer's impact on inference.
• Create enterprise knowledge graph datasets and evaluation methodologies.
• Collaborate with hardware/systems teams to keep software aligned with capability.
• Generate evidence to support pilots, fundraising, and technical validation.

Skills

Python

LLM fine-tuning

RAG/GraphRAG

Knowledge graphs

Performance engineering

Distributed systems

Benchmark frameworks

Communication

Tools

PyTorch

TensorFlow

Graph databases

GPU profiling tools

Job description

IC Resources is recruiting a Machine Learning Engineer for an Edinburgh-based AI startup. The role blends software modelling with benchmarking to validate system behaviour in a hybrid, UK-wide context.

You will develop agentic AI workloads, build a robust benchmark suite, and design experiments to measure how the semantic memory layer affects inference performance. Strong Python and distributed-systems know-how are essential.