XCRS
XCRS (Explainable Career, Course and Concept Recommendation System) suggests software roles that fit what you already know and what you're curious about, and tells you why. You sort skills into four boxes (enjoyed, neutral, didn't enjoy, curious about) or import them from a CV or LinkedIn PDF. It answers with up to three roles your skills support, your likely level in each, job titles to search for, and free resources for the first gaps.
Recommending and explaining are two separate steps. A scoring engine ranks 30 roles against a catalog of 259 skills checked against O*NET and ESCO, and its score mixes what you enjoy with what you already cover. Then a local language model writes each explanation from only the facts the engine used, so the reasons can't drift from the result. All the models are self-hosted and run through Ollama, with a copy on AWS.
It started as my master's thesis at Boğaziçi University and was published at ISD 2025 (paper). That research prototype compared five embedding providers and used GPT-4o for explanations. The version live today is a rebuild that follows what the prototype's user study and an expert evaluation found.