Offshore Wind RAG System

Built during my internship on Equinor’s Renewables Bid Excellence team, which works on the non-price criteria for offshore wind farm bids. Those bids turn on evidence spread across thousands of documents, and finding the right one was costing people hours.

• Retrieval. NLP to identify the genuinely relevant documents for a question out of thousands, rather than keyword matching and hoping.
• Generation. An LLM composes an answer grounded in the retrieved documents, with its sources attached. Non-negotiable in a bid context, where an uncited answer is unusable because someone has to stand behind the claim.
• Ingestion. A Selenium scraper that dynamically loads pages and extracts content to build the document database, since much of the source material was not sitting in a tidy indexed repository.
• Deployment. Docker and Kubernetes, behind a chatbot-style interface so non-technical bid teams could use it.

Search went from hours to seconds. The citation requirement ended up shaping the retrieval architecture more than any benchmark did.