Source description
About the role
Responsibilities - Drive R&D; and engineering for AI Infrastructure optimization within CloudKeeper's FinOps for AI platform - building the Tuner AI / Commit AI capability on GPU and ML workloads. - Design and build optimization engines for GPU right-sizing, idle shutdown, spot migration with checkpoint/resume automation, inference batching, quantization, and model placement. - Extend the optimization stack to LLM-era workloads - caching, model routing, dynamic batching, prompt optimization, RAG-aware architectures. - Partner with the Lens AI team to translate GPU and ML workload signals into actionable, dollar-quantified optimization recommendations for customers. - Work cross-functionally with product, platform, and customer success teams to ship optimization features end-to-end (data ingestion - optimization engine - customer-facing recommendation). - Lead technical direction for AI workload optimization, set engineering standards, and mentor the ML / MLOps engineering bench as the AI Infrastructure pillar scales. - (Lead level) Hire, ramp, and grow a team of ML infrastructure engineers as headcount expands.Must Have : - B.E / B.Tech / M.Tech / MCA with 7 years of hands-on engineering experience. - Production experience with GPU workloads - has measurably optimized GPU utilization, throughput, or cost in a real production environment, not just academic / lab work. - Strong performance engineering background - must come ready with a concrete optimization story including before/after metrics (latency, throughput, or cost reduction). - Solid Python Linux systems fundamentals. - Solid understanding of the ML model lifecycle - training, serving, inference - able to reason about what is running on the GPU and why. - MLOps fluency - model deployment, monitoring, observability, GPU cluster operations. - Hands-on with cloud GPU instances (AWS P5 / G6, Azure ND series, GCP A3, or equivalent) and Kubernetes-based GPU orchestration (EKS / AKS / GKE GPU node pools, Karpenter, Run:ai, NVIDIA GPU Operator, or similar). - Familiarity with at least one modern LLM inference framework - vLLM, TGI, Triton, SGLang, Ray Serve, or BentoML. - Strong communication skills - able to translate deep technical optimization into customer / business outcomes. - (Lead level) Experience managing or technically leading a team of 3 engineers.Good to Have : - Deep LLM-era optimization expertise - KV caching, semantic caching, model routing, dynamic batching, quantization (FP16 - INT8 - INT4), model distillation, structured outputs. - Familiarity with LLM workload patterns - RAG, agents, embeddings, vector databases (Pinecone, Weaviate, Qdrant). - CUDA, NCCL, mixed-precision training and inference. - Experience with managed ML training platforms - SageMaker, Azure ML, Vertex AI, Databricks Mosaic. - Exposure to GPU-native clouds - CoreWeave, Lambda Labs, RunPod, Crusoe. - Open source contributions to ML infrastructure projects - vLLM, llama.cpp, TGI, Ray, Triton, KubeRay. - Adjacent experience in cloud cost optimization / FinOps - Spot.io, ScaleOps, Granulate, CAST AI. - Comfort with Agile methodology and modern engineering practices (CI/CD, code review, observability). (ref:hirist.tech) . .