Lead NLP/LLM Engineer

Must-Have Skills:

LLM, MLOps, PyTorch, RAG, Vector Database, Cloud Server (Google / AWS), Python

Good-to-Have Skills:

CI/CD, GCP Vertex, LangChain, SageMaker, TensorFlow

About the Role:

Pentimenti AI (one of Uplers clients) is looking for a Senior Machine Learning Engineer who is passionate about their work, eager to grow, and committed to delivering exceptional results. If you are a team player with a positive attitude, this is the opportunity for you!

Agentic platforms represent the third wave of AI, enabling complex multi-step work via autonomous LLM-powered agents. You'll own the stackfrom research to deploymenthelping ship magical, high-impact features for users.

Key Responsibilities:

• Own the agentic & RAG roadmap: design, prototype, and launch LLM agents (planner-executor, multi-agent, tool-calling) with sub-second latency.
• Productionize RAG pipelines: embedding strategy, vector-DB design (Weaviate, Pinecone), hybrid search, evaluations, and guardrails.
• Fine-tune models with PEFT/LoRA, RLHF, and safety alignment; publish impactful research.
• Optimize inference: quantization (INT4/8), speculative decoding, TensorRT-LLM/vLLM, or Ray Serve to reduce token costs.
• Lead and mentor a high-agency team; establish MLOps, CI/CD, observability, and governance standards.
• Partner with product & design to turn research into scalable, user-facing features.

Core Qualifications:

• Experience: 5+ years in software/ML, including 2+ years in LLM/NLP product delivery.
• Deep Learning Stack: Python and PyTorch (TensorFlow/JAX welcome). CUDA/Triton knowledge is a plus.
• Agentic & RAG Frameworks: Experience with LangChain, LlamaIndex, CrewAI, and vector DBs like Weaviate, Pinecone, Qdrant.
• Model Optimization: Quantization, distillation, AWS Neuron, GPU kernel tuning.
• Cloud & MLOps: Kubernetes, Ray, SageMaker, or GCP Vertex. Familiar with Terraform/Pulumi and observability tools.
• Communication & Leadership: Strong design documentation and cross-functional leadership skills.

Bonus Skills:

• Multimodal agent systems (vision-language, audio-language)
• Privacy-preserving ML (federated learning, differential privacy)
• Open-source contributions (LangChain, Pinecone, Triton, etc.)