Ask MARIC
Ask a question of the last 90 days of rail news
Answers are grounded in items we have read and cite them as [1], [2]… Where the feed is silent, the assistant says so. English or Bahasa Malaysia.
thinking — searching the library and drafting an answer…
How it works
1. Your question is matched against every enriched item with hybrid search: keyword (BM25) plus semantic embeddings, fused by rank.
2. The top items are passed to the model with a strict instruction: answer only from these, cite each claim, and name the clusters and members who may care.
3. Nothing you type is used to train any model. Usage is capped per day to keep cost predictable.
Production path: the same pattern over MARIC's own documents — handbooks, standards, member capability statements — with role-based access.