Data Scientist - LLM Engineering

About The Opportunity :
We are a fast-growing enterprise AI & data science consultancy serving global clients across finance, healthcare, and enterprise software.The team builds production-grade LLM-driven productsRAG systems, intelligent assistants, and custom inference pipelinesthat deliver measurable business outcomes.Role & Responsibilities :
- Design, fine-tune and productionize large language models (instruction tuning, LoRA/PEFT) using PyTorch and Hugging Face tooling for real-world applications.- Architect and implement RAG pipelines : embeddings generation, chunking strategies, vector search integration (FAISS/Pinecone/Milvus) and relevance tuning for high-quality retrieval.- Build scalable inference services and APIs (FastAPI/Falcon), containerize (Docker) and deploy to cloud/Kubernetes with low-latency and cost-optimized inference (quantization, ONNX/Triton).- Collaborate with data engineers and ML scientists to productionize data pipelines, automate retraining, monitoring, evaluation and drift detection.- Drive prompt-engineering, evaluation frameworks and safety/guardrail implementation to ensure reliable, explainable LLM behavior in production.- Establish engineering best-practices (Git workflows, CI/CD, unit tests, observability) and mentor junior engineers to raise team delivery standards.Skills & Qualifications :
Must-Have :
- 4+ years in data science/ML engineering with demonstrable experience building and shipping LLM-based solutions to production.- Strong Python engineering background and hands-on experience with PyTorch and Hugging Face Transformers (fine-tuning, tokenizers, model optimization).- Practical experience implementing RAG : embeddings, vector DBs (FAISS/Pinecone/Weaviate/Milvus), chunking and retrieval tuning.- Production deployment experience : Docker, Kubernetes, cloud infra (AWS/GCP/Azure) and inference optimization (quantization, batching, ONNX/Triton).Preferred :
- Experience with LangChain/LangGraph or similar orchestration frameworks, and building agentic workflows.- Familiarity with ML observability, model governance, safety/bias mitigation techniques and cost/performance trade-offs for production LLMs.