Spaces:
Sleeping
Sleeping
Commit History
-
feat: college keyword lookup, smart result count, field filtering, context window optimisation
- build_college_keyword_map(): builds normalised keyword→college map from CollegeNameSearchKeyword1-4 columns (5,271 keys, 2,505 colleges) - resolve_college_in_query(): resolves abbreviations like 'M N College'→ 'M. N. COLLEGE, VISNAGAR', 'IMN Law'→'I. M. Nanavati Law College' - Noise guards: stopwords + Gujarat cities prevent false matches on generic words - Plausibility: allows 2+ single-letter initials (m, n) + 3-5 char abbreviations - CRITICAL FIX: load_cache() now calls build_college_keyword_map() so keyword map is populated on server startup (was only building vocabulary) - search_engine: _dedup_by_college() collapses same college to one result - search_engine: _smart_cutoff() returns 1 result for high-score specific queries, uses gap-based pruning to avoid padding with weakly related results - search_engine: early intent detection before FAISS for correct table routing - search_engine: larger FAISS pool (10x top_k) for post-filtered intents - field_filter: detect_intent(…b26d5a9 -
initial deploy
6fc17e7
