End-to-end AI platform work — not an API call to a model provider behind a chat widget. Data pipelines, model selection or fine-tuning, evaluation, and the MLOps to keep it honest in production.
What the work involves
Models & fine-tuning
Foundation model selection (open + closed weights) by task fit and cost
Fine-tuning, LoRA, and instruction-tuning where it earns its keep
Classical ML and neural-net models when LLMs are the wrong tool
Pipelines & RAG
Data prep, embedding, and vector store design
Retrieval-augmented generation with chunking + reranking
Agent orchestration with tool use, memory, and guardrails
MLOps & evaluation
Eval harnesses with regression tests on prompt + model changes
Tracing, observability, and cost attribution per request
