LLM Inference Performance Engineer

Bitdeer (NASDAQ: BTDR); Singapore; On-site; SGD 120,000 - 180,000; Full time

Job summary

Bitdeer in Singapore seeks an Inference Performance Engineer to calibrate real production traffic shapes and make benchmarking reflect actual usage, aligning GPU fleets and cost models with live telemetry.

The role blends data science, systems thinking, and collaboration with infrastructure, finance, and engineering teams to produce defensible capacity forecasts and cost-per-token metrics for the MaaS platform.

Qualifications

• 5+ years of experience in performance engineering, capacity planning, or applied ML systems on large-scale services.
• Fluent in profiling production traffic and building distributions of request rate, token lengths, concurrency, and burstiness.
• Experience with Python/pandas, SQL, Spark or similar data-analysis stack.
• Knowledge of LLM inference internals (batching, KV-cache, prefill/decode) and serving frameworks.
• Experience designing or auditing load-testing/benchmarking tooling and reconciling against live telemetry.

Responsibilities

• Traffic and Workload Characterization: profile production streams to model prompt/output lengths, concurrency, and burstiness by model and customer segment.
• Harness-to-Production Fidelity: validate load-testing against live telemetry and adjust benchmarks accordingly.
• Capacity Planning and Unit Economics: translate workload data into cost-per-token/ per-request models and inform fleet sizing and pricing.
• Metric Definition Ownership: define latency percentiles, throughput, and utilization metrics and ensure consistent definitions across teams.
• Cross-Team Performance Reporting: maintain dashboards and postmortems to provide a trusted view of inference performance and cost trends.

Skills

Performance engineering

Capacity planning

ML systems

Python

SQL

Spark

Tools

Prometheus

Grafana

vLLM

TensorRT-LLM

SGLang

Triton Inference Server

Job description

Bitdeer in Singapore seeks an Inference Performance Engineer to calibrate real production traffic shapes and make benchmarking reflect actual usage, aligning GPU fleets and cost models with live telemetry.

The role blends data science, systems thinking, and collaboration with infrastructure, finance, and engineering teams to produce defensible capacity forecasts and cost-per-token metrics for the MaaS platform.