AI/ Generative AI Engineer

Role & responsibilities

Design, fine-tune, and evaluate generative models (transformers) for tasks such as summarization, Q&A, code generation, and retrieval-augmented generation (RAG).

• Implement data pipelines for training & evaluation: dataset collection, cleaning, labeling, and augmentation.
• Develop, test, and maintain prompt engineering practices and templates; measure prompt drift and performance.
• Build RAG pipelines (embeddings, vector store selection, index management, retriever tuning).
• Containerize models and services (Docker), create reproducible deployments (FastAPI / Flask / .NET wrappers), and help deploy to staging/production (K8s, serverless, or cloud infra).
• Implement monitoring, logging, and evaluation metrics for model performance and data/feature drift.
• Work with product and infra teams to integrate AI features into user-facing apps and ensure secure usage (rate-limits, content filtering, PII redaction).
• Keep up with new model releases and evaluate third-party APIs (OpenAI, Anthropic, Meta, etc.) for integration.
• Write clear documentation, runbooks, and reproducible experiments.

Required qualifications

• 23 years professional experience in applied ML / NLP / generative model work.
• Strong Python skills and experience with ML frameworks: PyTorch (preferred) or TensorFlow.
• Experience with transformer models and libraries: Hugging Face Transformers, sentence-transformers, or equivalent.
• Experience with embeddings and vector DBs (e.g., FAISS, Milvus, Pinecone, Weaviate).
• Good understanding of model evaluation: ROUGE, BLEU, Accuracy, F1, human eval basics, and safety metrics.
• Solid software engineering fundamentals: Git, unit testing, code reviews, and RESTful APIs.
• Experience with LLM orchestration tools / agent frameworks (LangChain, LlamaIndex, LangGraph, Semantic Kernel/Autogen).

Deliverables / KPIs (first 36 months)

• Ship at least one end-to-end GenAI feature (prototype staged deployment) with documented evaluation results.
• Reproducible training/fine-tuning pipeline and an experiment tracking dashboard.
• Production-ready inference endpoint with basic monitoring and cost controls.
• Documented prompt templates and a rollback strategy for model releases.