Lead TPU Scheduling & Optimization Architect

EPAM Systems Inc; United States; Remote; USD 130,000 - 190,000; Full time

Job summary

EPAM Systems is seeking a Senior/Lead Optimization/Algorithms Engineer to design and build the combinatorial optimization core powering topology-aware TPU scheduling. You will translate scheduling constraints into solvable mathematical models and drive sub-second performance at scale.

Responsibilities include developing a solver core for TPU slice placement, building MIP/CP-SAT representations of multi-host constraints, profiling latency, and integrating Kubernetes scheduling considerations into

Qualifications

• Strong background in combinatorial optimization, operations research, or algorithms engineering.
• Experience with MIP/CP-SAT solvers (OR-Tools, Gurobi, CPLEX).
• Proficiency in Python and/or Golang.
• Understanding of graph theory and constraint satisfaction problems.
• Familiarity with performance profiling and latency-critical systems.

Responsibilities

• Develop the combinatorial solver core for TPU slice placement.
• Build MIP/CP-SAT models representing multi-host slice constraints.
• Perform latency profiling to ensure sub-second solve times at scale.
• Translate Kubernetes CRD/Pod spec constraints into abstract graph/optimization models.
• Collaborate with AI/ML Performance Engineers to incorporate empirical hardware coefficients into the cost model.

Skills

Combinatorial optimization

Operations research

Algorithms engineering

Python

Golang

Graph theory

Latency profiling

Tools

OR-Tools

Gurobi

CPLEX

Job description

EPAM Systems is seeking a Senior/Lead Optimization/Algorithms Engineer to design and build the combinatorial optimization core powering topology-aware TPU scheduling. You will translate scheduling constraints into solvable mathematical models and drive sub-second performance at scale.

Responsibilities include developing a solver core for TPU slice placement, building MIP/CP-SAT representations of multi-host constraints, profiling latency, and integrating Kubernetes scheduling considerations into