Role: Senior AI / ML Architect
Location: SFO, CA (Onsite)
Employment Type: Full-Time with Criticalriver
Experience: 10–15 Years (4+ years ML/DL, 2+ years LLM/GenAI)
Job Summary
We are seeking a highly skilled Senior AI / ML Architect with strong hands-on experience in LLM training, fine-tuning, cloud deployment, model hosting, and MLOps. The ideal candidate will lead the end-to-end lifecycle of AI models, from data preparation and model selection to deployment, monitoring, and optimization in production environments.
Key Responsibilities
- Design and implement AI/ML solutions for training and fine-tuning LLMs using enterprise data.
- Build end-to-end pipelines covering data preparation, model training, evaluation, deployment, and monitoring.
- Perform hands-on LoRA/QLoRA fine-tuning and optimize training parameters.
- Develop scalable GPU-based training environments on AWS, Azure, or Google Cloud Platform.
- Deploy and host models using vLLM, TensorRT-LLM, SGLang, or TGI.
- Establish MLOps pipelines using MLflow, W&B, Docker, Kubernetes, Ray, and CI/CD tools.
- Monitor model performance, infrastructure utilization, latency, throughput, and costs.
- Troubleshoot training, inference, deployment, and production issues.
- Provide technical leadership and mentor AI/ML engineering teams. Mandatory Skills
1. LLM Training & Fine-Tuning
- PyTorch, Hugging Face Transformers
- LoRA, QLoRA, PEFT, TRL, Accelerate
- Experience fine-tuning LLMs such as Llama, Qwen, Mistral/Mixtral, Gemma, DeepSeek, or Phi
- Strong understanding of datasets, hyperparameters, and GPU optimization
2. Model Hosting & Cloud Deployment
- AWS, Azure, or Google Cloud Platform
- GPU infrastructure (A100, H100, L40S)
- Docker, Kubernetes
- vLLM, SGLang, TensorRT-LLM, TGI
- Production-grade model serving architectures
3. MLOps & Lifecycle Management
- MLflow, W&B
- Model registry, versioning, monitoring, rollback
- CI/CD for ML pipelines
- Automated training, deployment, and retraining workflows
Preferred Skills
- Python, SQL
- Data engineering and large-scale data pipelines
- Spark, Databricks
- Quantization techniques (GPTQ, AWQ, bitsandbytes, FP8)
- RAG, Vector Databases, Agentic AI, MCP, Tool Calling
- LLM evaluation and regression testing frameworks
