AI/ML Technical Leader - Language Model Inference & AI Ops

The application window is expected to close on: 09/30/2026 Job posting may be removed earlier if the position is filled or if a sufficient number of applications are received . This is a HYBRID role in San Jose, CA. Must be able to work on site 3 days per week. Meet the Team   Join Cisco’s CX AI Incubation Team as an AI Operations Technical Leader and help productionize LLM/SLM capabilities for Intelligent Customer Experiences, across cloud and on-prem environments. In Cisco CX, you will build and operate scalable AI systems that move from prototype to production, powering delivery intelligence, network automation, infrastructure testing, and intelligence on edge.  You will collaborate with product and engineering teams to deploy reliable, secure, and observable AI services, optimizing inference performance from CPU and small GPUs to large multi-GPU servers, including air-gapped and customer-managed deployments.  You'll work on cutting-edge inference optimization - speculative decoding, continuous batching, quantization, and KV-cache strategies to deliver cost-effective, low-latency AI across cloud, on-prem, and air-gapped environments. Your Impact   Join Cisco’s Customer Experience (CX) AI Incubation team to build and run production-grade AI platforms and services that transform customer engagement and operational efficiency. You will focus on end-to-end AI DevOps for LLMs/SLMs, including on-prem inference packaging, runtime optimization, deployment automation, and model/service observability. This role requires strong software engineering, hands-on GPU inference experience, and a track record of operationalizing models at scale.  What You’ll Do   Productionize LLM/SLM-powered features by building robust model-serving and deployment pipelines (cloud + on-prem) with clear SLAs, monitoring, and rollback strategies. Optimize inference performance across CPUs and small GPUs using techniques such as speculative decoding, continuous batching, paged attention, KV-cache reuse, and low-bit quantization (F8/INT4) for cost and latency wins. Package and integrate on-prem inference stacks (VM/containers) with customer environments, including secure configuration, versioning, and upgrade-safe deployments.  Design scalable serving architectures for generative AI (multi-tenant, secure, cost-aware), including tensor/pipeline parallelism, disaggregated prefill/decode, capacity planning, and performance benchmarking. Build automated CI/CD for models and prompts: evaluation gates, regression testing, artifact management, and reproducible releases.  Implement model and service observability: latency/throughput metrics, quality drift signals, safety checks, and incident triage workflows.  Support training and fine-tuning workflows for LLMs/SLMs, including data curation, experiment tracking, and packaging models for production. Partner with product and engineering to integrate AI services into applications, ensuring reliability, security, and responsible AI behavior. Evaluate and adopt emerging inference techniques and runtimes; drive build-vs-adopt decisions across vLLM, TensorRT-LLM, SGLang, llama.cpp, and similar engines based on workload characteristics. Minimum Qualifications   Bachelor’s degree with 9+ years of related experience, or Masters degree with 7+ years of related experience.  Experience in Python, Java or C++, and building production services for ML/AI workloads.  Experience with PyTorch/TensorFlow and tooling across the ML lifecycle (data pipelines, training, evaluation, deployment).  Experience deploying and operating NLP/Generative AI systems in production, including performance tuning and reliability practices.  Experience working in cross-functional teams, delivering in fast-paced environments, and communicating technical concepts clearly.  Preferred Qualifications   ~Inference & Serving Proven experience productionizing LLMs/SLMs with GPU-backed inference and runtime optimization. Hands-on experience with inference engines – vLLM, TensorRT-LLM, Triton, SGLang, llama.cpp and GPU profiling (Nsight, PyTorch profiler). Working knowledge of speculative/assisted decoding, continuous batching, paged/flash attention, KV-cache management, and structured/constrained decoding (guided JSON, grammar-based). Experience with quantization techniques (GPTQ, AWQ, SmoothQuant, FP8, INT4) and accuracy/perf tradeoffs. Familiarity with multi-GPU parallelism (tensor, pipeline, expert) and disaggregated serving patterns.  ~Model Adaptation Experience with PEFT (LoRA, QLoRA), distillation, and SLM specialization for domain-specific deployments. Familiarity with LLM-evaluation (LLM-as-a-judge, golden sets, drift detection, regression gates). ~On-Prem, Edge & Infra Hands-on experience with on-prem deployment patterns (air-gapped, customer-managed), packaging, integration, upgrade strategy. Exposure to edge/resource-constrained inference (CPU, NPU, small GPU; runtimes like llama.cpp, ONNX Runtime, OpenVINO, MLC). Experience with AI infra and MLOps tooling such as K8s, CI/CD, model registry, experiment tracking, observability. ~Communication Strong written and verbal communication; ability to drive design reviews and produce clear technical documentation. Why Cisco?  At Cisco, we’re revolutionizing how data and infrastructure connect and protect organizations in the AI era – and beyond. We’ve been innovating fearlessly for 40 years to create solutions that power how humans and technology work together across the physical and digital worlds. These solutions provide customers with unparalleled security, visibility, and insights across the entire digital footprint. Fueled by the depth and breadth of our technology, we experiment and create meaningful solutions. Add to that our worldwide network of doers and experts, and you’ll see that the opportunities to grow and build are limitless. We work as a team, collaborating with empathy to make really big things happen on a global