Callosum

Callosum is inviting exceptional students to express interest in future internships working on AI infrastructure, heterogeneous compute, and advanced accelerator systems. Internships are arranged on an ongoing, needs-based basis rather than through a formal fixed programme.

Build the foundational runtime for long-running AI agents operating across distributed infrastructure, with durable execution, secure tool use, scheduling, and deep observability. This high-leverage role combines distributed systems, framework design, and runtime engineering from first principles.

Callosum is seeking an ML Research Engineer to investigate why agentic AI systems fail and build evidence-driven methods for improving their intelligence, cost, and reliability. The role combines independent research, experimental engineering, agent evaluation infrastructure, and publication in a London-based team working across heterogeneous AI hardware.

Own and advance Callosum’s LLM-guided program evolution and optimisation system, enabling automated search across models, software, hardware, and workloads. This research-heavy engineering role combines evolutionary optimisation, search strategy design, and reliable internal infrastructure.

Own and build Callosum’s rigorous benchmarking system for agentic and algorithmic LLM approaches, using reproducible execution-based evaluation rather than self-reported scores. The role combines research-grade evaluation with hands-on engineering and directly informs product decisions, customer proof points, and published benchmarks.

Own the communication systems that let heterogeneous AI accelerators operate as one efficient system. You’ll improve data movement, networking, memory transfer, and interconnect performance across emerging hardware platforms.

Own inference performance and deployment for heterogeneous AI infrastructure, running production-equivalent experiments across cloud and on-prem hardware. You’ll build reproducible deployment patterns, benchmarking systems, and orchestration improvements that guide the stack toward production readiness.

Build the next generation of heterogeneous AI inference engines by extending systems such as vLLM and SGLang for hardware-aware scheduling, memory management, execution, and serving. You’ll design production-scale parallelism and disaggregation strategies across diverse accelerators.

Build the low-level systems software that enables AI inference across heterogeneous accelerators and frontier silicon. You’ll develop kernels, drivers, firmware, runtimes, and scheduling systems that optimize performance and reliability across diverse hardware.

Callosum is seeking expressions of interest from exceptional engineers to build AI infrastructure spanning heterogeneous accelerators, intelligent orchestration, and frontier silicon. The work covers high-performance systems, inference infrastructure, hardware bring-up, accelerator programming, simulation, and distributed systems.