ML Engineer - Inference

📋 Description

• • Shape the public face of Mindbeam’s next-generation AI infrastructure by designing and owning the inference APIs, SDKs, and command-line tools that thousands of researchers, developers, and Fortune-500 engineers will rely on every day.
• • Translate cutting-edge research models—from billion-parameter transformers to novel sparse architectures—into rock-solid, low-latency production endpoints that serve millions of predictions per second with sub-100 ms p99 latency.
• • Build intuitive abstractions that hide the complexity of distributed inference, automatic batching, dynamic quantization, and hardware-specific optimization (GPU, TPU, Inferentia) behind clean, idiomatic Python, REST, and gRPC interfaces.
• • Partner shoulder-to-shoulder with our research team to co-design new model-serving paradigms, then shepherd those ideas through alpha, beta, and GA by writing design docs, RFCs, and example notebooks that turn bleeding-edge science into copy-paste developer joy.
• • Own the full lifecycle of our inference stack: profiling, benchmarking, autoscaling policies, canary rollouts, and real-time monitoring with Prometheus, Grafana, and custom ML health checks to guarantee five-nines reliability for enterprise SLAs.
• • Champion the developer experience: run weekly user interviews, mine GitHub issues, and instrument SDK telemetry to discover friction points, then ship weekly releases that cut integration time from hours to minutes.
• • Craft security-first architectures that satisfy SOC 2, HIPAA, and FedRAMP controls—end-to-end TLS, customer-managed keys, VPC peering, and fine-grained IAM—so regulated industries can adopt Mindbeam without a second thought.
• • Contribute to open-source: upstream improvements to TensorRT, vLLM, and Kubernetes Serving, publish blog posts and conference talks, and grow a vibrant community that extends our platform in ways we never imagined.
• • Mentor junior engineers through pair programming, design reviews, and lightning talks; foster a culture where curiosity, psychological safety, and constructive dissent lead to breakthrough ideas.
• • Iterate at startup speed: ship an MVP in days, measure, learn, and pivot without ego, while still building the robust foundations that will scale to exaFLOP clusters tomorrow.