# Quickstart: Semantic Search on Multimodal & Video Foundation Models (Universal V15.0)
import pyarrow.parquet as pq
import numpy as np

# 1. Load Parquet Dataset
table = pq.read_table("MULTIMODAL_VISION_LANGUAGE_VIDEO_FOUNDATION_MODELS_2026_FULL.parquet")
df = table.to_pandas()

print(f"Loaded {len(df)} Multimodal Vision-Language & Video Foundation Model research papers.")
print(f"Sample Top Paper: {df['title'].iloc[0]} (Citations: {df['academic_citations_count'].iloc[0]} | Stars: {df['github_stars'].iloc[0]})")
print(f"Vision Backbone: {df['vision_backbone_architecture'].iloc[0]}")
print(f"Modalities: {df['supported_modalities'].iloc[0]}")
print(f"IP Safety: {df['commercial_ip_verdict'].iloc[0]} (Score: {df['commercial_ip_safety_score'].iloc[0]}/100)")

# 2. Example Semantic Vector Search
query_vector = np.random.randn(384).astype(np.float32)
query_vector /= np.linalg.norm(query_vector)

abstract_vectors = np.vstack(df['abstract_vector_384d'].values)
similarities = np.dot(abstract_vectors, query_vector)
top_5_idx = np.argsort(similarities)[::-1][:5]

print("\n--- TOP 5 MULTIMODAL SEMANTIC VECTOR SEARCH RESULTS ---")
for idx in top_5_idx:
    print(f"Score: {similarities[idx]:.4f} | {df['title'].iloc[idx]} (Citations: {df['academic_citations_count'].iloc[idx]} | Backbone: {df['vision_backbone_architecture'].iloc[idx]})")