Inference Performance Engineer - MaaS
Bitdeer is a technology company providing Bitcoin mining solutions.
Funding history
Investors
Projects
About Bitdeer
Bitdeer provides full-spectrum Bitcoin mining and high-performance computing solutions, including SEALMINER mining equipment, Minerbase cooling containers, cloud mining, co-mining, mining management applications, mining rights marketplaces, and large-scale data center operations. The company also offers AI cloud infrastructure with GPU computing, model training and deployment capabilities, and turnkey AI data center solutions for enterprise customers and developers. Bitdeer is headquartered in Singapore and operates globally.
Skills
About the Role
You will profile production inference traffic, validate benchmark workloads against telemetry, and create capacity and unit-economics models. You will define performance metrics, maintain dashboards and reports, reconcile costs with actual spend, and help teams make reliable performance and pricing decisions.
Requirements
- 5+ years of performance engineering, capacity planning, or applied ML systems experience
- Experience with GPU-backed or latency-sensitive distributed services
- Experience with LLM or ML inference workloads
- Experience analyzing production traffic with Python, pandas, SQL, Spark, or comparable tools
- Knowledge of LLM inference internals including continuous batching, KV-cache management, and prefill/decode scheduling
- Experience with load-testing or benchmarking tools
- Experience with Prometheus, Grafana, and distributed tracing
Responsibilities
- Profile production traffic across chat, completion, embedding, and fine-tuning endpoints
- Build and maintain statistical workload models and traffic profiles
- Validate benchmarking and load-testing harnesses against production behavior
- Create capacity forecasts and cost-per-token or cost-per-request models
- Define canonical latency, throughput, and utilization metrics
- Document instrumentation and aggregation requirements
- Build dashboards, reports, and postmortem inputs for performance and cost trends