Job Description & Details
This is a heavy-hitting technical lead role where you aren't just writing code in a silo, but also sitting in front of enterprise customers translating messy business problems into production AI systems. If you enjoy building end-to-end solutions involving foundation models, computer vision, and agentic workflows while mentoring engineers, this is a great match.
What You'll Actually Be Doing
Your days will be split between deep engineering and high-stakes technical consulting. You'll architect multi-agent systems using frameworks like LangGraph and Claude, design computer vision pipelines for unstructured data, and handle forecasting models. Half the time you'll be writing clean PyTorch code and optimizing inference with vLLM, and the other half you'll be whiteboard sessions with clients, breaking down ambiguous requirements into actionable technical milestones.
The Core Tech Stack
You need a battle-tested command of PyTorch, Hugging Face, and PEFT for model fine-tuning, alongside serving engines like vLLM and Ollama. Because this role leans heavily into modern GenAI, deep fluency in agentic frameworks like LangGraph, AutoGen, and the Model Context Protocol (MCP) is non-negotiable. Combined with solid MLOps tooling (Docker, Kubernetes, MLflow) and traditional computer vision libraries, you'll need to bridge the gap between low-level model optimization and high-level system architecture.
Interview Expectations
Expect the hiring team to grill you on inference optimization and scaling tradeoffs, likely asking how you would handle KV-cache bottlenecks and quantization strategies when deploying large foundation models like Claude or Llama variants under strict latency constraints. They'll also test your system design chops by asking you to whiteboard a multi-agent tool-using workflow, looking specifically at how you implement memory persistence, failure recovery loops, and guardrails to prevent agent hallucination in production.
Application Advice
Skip the generic resume fluff and directly showcase your war stories with numbers—highlight the 10+ production models you've shipped and explicitly mention your hands-on work with quantization, vLLM, and agentic frameworks like LangGraph or CrewAI. Ensure keywords like 'agentic workflows', 'fine-tuning (LoRA/QLoRA)', and 'PyTorch' are prominently featured in your experience bullets to clear the ATS and catch the eye of engineering leadership.