Agentic Compiler Engineer for GPU-Accelerated LLMs

Kog; Paris; On-site; EUR 90,000 - 130,000; Full time

Benefits offered by this job

Work across stack

Direct access to engineers across full

Fast optimization loop

Deep technical focus

High ownership

Job summary

Kog is building AGCO, a next‑gen agentic compiler for real‑time LLM optimization on datacenter GPUs. You will contribute to AGCO by exploring, compiling, verifying, and measuring optimizations across the stack on real hardware, collaborating with researchers and engineers.

This Paris‑based role focuses on IR design, compiler passes, GPU execution, and LLM inference optimizations, with ownership over critical decisions and the chance to push the boundaries of LLM performance.

Qualifications

• Experience in compiler engineering, including optimization passes, IRs, lowering, code generation, LLVM, or MLIR.
• Experience GPU programming with CUDA, HIP, Metal, Vulkan, or similar technologies.
• GPU performance work involving kernels, memory, synchronization, profiling, or hardware behavior.
• LLM inference engines and performance optimization.
• Attention, MoE, parallelism, communication, or other systems‑level parts of LLM execution.
• Formal verification, equivalence checking, SAT/SMT, or related methods.
• Systems that generate, search, test, benchmark, or optimize code automatically.

Responsibilities

• Compiler and IR design for representing and transforming LLM computations.
• Optimization passes, lowering, and code generation.
• Search methods for exploring different implementations and execution strategies.
• Verification and correctness checks for generated changes.
• GPU execution, profiling, and performance optimization.
• LLM inference across operators, memory, parallelism, and communication.
• Optimization loops that connect generated changes to measurements on real GPUs.

Skills

Compiler engineering

GPU programming

GPU performance

LLM inference engines

Attention / MoE / parallelism

Formal verification

Code generation automation

Tools

LLVM

MLIR

Job description

Kog is building AGCO, a next‑gen agentic compiler for real‑time LLM optimization on datacenter GPUs. You will contribute to AGCO by exploring, compiling, verifying, and measuring optimizations across the stack on real hardware, collaborating with researchers and engineers.

This Paris‑based role focuses on IR design, compiler passes, GPU execution, and LLM inference optimizations, with ownership over critical decisions and the chance to push the boundaries of LLM performance.