CENTRAL QUESTION
AI AGENTS
Build the loop from scratch, inspect graph state and traces, connect tools safely, and decide where autonomy must stop.
BOUNDED RANKING TOOL
Candidate generation, personalised scoring, constraint-aware re-ranking and evaluation belong to a dedicated system—not to unconstrained agent reasoning.
The agent may provide context and request a ranked set. The recommender enforces eligibility and safety, explains score components and may abstain. The agent cannot override eligibility, lower safety thresholds or turn a ranking into scientific evidence.
Open Recommender Systems Lab