""" ZKCoding Knowledge + Embedding 集成实验 使用原项目的 KnowledgeCatalog 和 SQLite Repository, 然后调用自己编写的 KnowledgeReranker。 运行: python -m experiments.knowledge_integration_test """ import tempfile from pathlib import Path from src.adapters.indexes.knowledge_sqlite import ( SqliteKnowledgeRepository, ) from src.modules.knowledge.application.catalog import ( KnowledgeCatalog, ) from src.modules.skill.domain.models import Actor from experiments.knowledge_reranker import KnowledgeReranker def main(): # 1. 创建测试用户 actor = Actor( tenant_id="test-tenant", user_id="developer", roles=[], ) # 2. 创建临时数据库 with tempfile.TemporaryDirectory() as temp_dir: folder = Path(temp_dir) repo = SqliteKnowledgeRepository( folder / "index.sqlite" ) # 3. 复用组长已有的测试数据生成函数 from src.modules.knowledge.tests.test_knowledge import ( write_fixture, ) corpus_dir = folder / "corpus" catalog_path = write_fixture(corpus_dir) repo.import_corpus( corpus_dir, actor.tenant_id, catalog_path, ) # 4. 创建真正的 Knowledge Service catalog = KnowledgeCatalog(repo) # 5. 调用原有 SQLite FTS 检索 query = "inventory" records = catalog.search( actor, query, ["fixture-java"], ) print("\n【SQLite 检索结果】") print("找到文档数量:", len(records)) for record in records: print( record["citation_id"], record["title"], record["source_url"], ) # 6. 调用我们新增的 Embedding 重排序 reranker = KnowledgeReranker() ranked = reranker.rerank( query, records, ) print("\n【Embedding 重排序结果】") for record in ranked: print( record["citation_id"], "相似度:", round(record["semantic_score"], 4), "来源:", record["source_url"], ) # 7. 验证引用信息没有丢失 assert len(ranked) == len(records) assert all( r["source_url"] for r in ranked ) print("\n集成测试通过:搜索与重排序已连接!") if __name__ == "__main__": main()