Enhancing Context-Aware Insight Generation in Corporate Learning Platforms via Retrieval-Augmented Language Models. University of Oulu, 2026.
My Master’s thesis, done at Valamis: a context-aware AI assistant for the company’s learning platform. It uses retrieval-augmented generation to ground an LLM in domain knowledge, so that automated reporting and business-intelligence queries over corporate learning and engagement data come back with answers that hold up.
The system
A hybrid retrieval pipeline feeds the assistant: deterministic column mapping, intent detection, and dense semantic search over a 37-document knowledge base, integrated into the platform’s existing OpenAI Assistants API workflow.
The evaluation
The part I care most about. I built two LLM-based evaluators, an answer-quality judge and a retrieval-relevance labeller, and validated both against human ratings before trusting either.
With that in place: overall answer quality reached 82.7%, a +13.6-point gain over the manual baseline, and the hardest business-strategy questions went from 34.7% to 76.7%. The deployed retrieval pipeline reached MRR 0.96 and Precision@1 0.93.
I delivered the validated system to Valamis’s development environment, working across AI, product and platform teams to align the outcomes with the product roadmap. The thesis was assessed at 5, on a scale where 5 is the highest grade.
The code is private under an NDA. The thesis itself is open access.