ML Ops Engineer — Agentic AI Lab (Founding Team)
Fabrion · San Francisco Bay Area
About this role
## ML Ops Engineer — Agentic AI Lab (Founding Team) **Location:** San Francisco Bay Area **Type:** Full-Time **Compensation:** Competitive salary + meaningful equity (founding tier) Backed by **8VC**, we’re building a world-class team to tackle one of the industry’s most critical infrastructure problems. --- ## About the Role Our AI Lab is pioneering intelligent infrastructure through **open-source LLMs**, **agent-native pipelines**, **RAG**, and **knowledge-graph-grounded models**. We’re hiring an **ML Ops Engineer** to connect ML research to production systems—owning automation for **training, deployment, versioning, and observability** pipelines that power our agents and AI data fabric. --- ## Responsibilities Build and maintain **secure, scalable, automated pipelines** for: - **LLM fine-tuning & training:** SFT, LoRA, RLHF, DPO - **RAG embedding pipelines** with dynamic updates - **Model conversion, quantization, and inference rollout** - **Hybrid compute infrastructure** (cloud, on-prem, GPU clusters) using **Kubernetes, Ray, Terraform** - **Containerization** with **Docker** + reproducible builds and **CI/CD** (GitHub Actions or ArgoCD) - **Model governance:** versioning, metadata, lineage, reproducibility, evaluation capture - **Evaluation & benchmarking frameworks** (e.g., OpenLLM-Evals, RAGAS, LangSmith) - **Security & access control integration** (OPA, ABAC, Keycloak) to enforce model policies per tenant - **Observability:** latency, token usage, performance metrics, error tracing, drift detection - **Agentic app deployment** using **LangGraph, LangChain**, and custom inference backends (vLLM, TGI, Triton) --- ## Desired Experience ### Model Infrastructure - **4+ years** in MLOps / ML platform engineering / infra-focused ML roles - Familiar with model lifecycle tools: **MLflow, Weights & Biases, DVC, HuggingFace Hub** - Experience deploying **open-source LLMs** (e.g., LLaMA, Mistral, Falcon, Mixtral) - Comfortable with tuning: **HuggingFace Trainer, DeepSpeed, FSDP, QLoRA** - Inference serving: **vLLM, TGI, Ray Serve, Triton** ### Automation + Infra - Proficient with **Terraform, Helm, Kubernetes**, and container orchestration - Experience with **ML CI/CD** (e.g., GitHub Actions + model checkpoints) - Managed hybrid GPU workloads (e.g., Lambda, Modal, HF Inference, SageMaker) - Cost optimization: spot scaling, batch prioritization, model sharding ### Agent + Data Pipeline Support - Familiar with **LangChain, LangGraph, LlamaIndex** (or similar RAG/agent orchestration) - Built embedding pipelines for multi-source documents (PDF/JSON/CSV/HTML) - Vector DB experience: **Weaviate, Qdrant, FAISS, Chroma** ### Security & Governance - Implemented model-level **RBAC**, usage tracking, audit trails - Integrated with rate limits, tenant billing, and SLA observability - Policy-as-code: **OPA / Rego** and access layers --- ## Preferred Stack - **LLM Ops:** HuggingFace, DeepSpeed, MLflow, Weights & Biases, DVC - **Infra:** Kubernetes (GKE/EKS), Ray, Terraform, Helm, GitHub Actions, ArgoCD - **Serving:** vLLM, TGI, Triton, Ray Serve - **Pipelines:** Prefect, Airflow, Dagster - **Monitoring:** Prometheus, Grafana, OpenTelemetry, LangSmith - **Security:** OPA (Rego), Keycloak, Vault - **Languages:** Python (primary), Bash; optionally Rust/Go --- ## Mindset & Culture Fit - Builder mindset with startup autonomy: **automate what slows you down** - Obsessive about **reproducibility, observability, traceability** - Comfortabl
Listing freshness
CronJobs last confirmed this listing 1h ago. If its source stops confirming the opening for seven days, this page is removed from active inventory.