Spaces:
Sleeping
Sleeping
Aissugur Kurganbekov
Claude Opus 4.8
Add hybrid retrieval (TF-IDF + embeddings), OCR, A/B eval, and answer trust/UX
f65a6b3 | """Database models: Document, Chunk, and QA history.""" | |
| from datetime import datetime, timezone | |
| from sqlalchemy import ( | |
| Column, | |
| DateTime, | |
| ForeignKey, | |
| Integer, | |
| LargeBinary, | |
| String, | |
| Text, | |
| ) | |
| from sqlalchemy.orm import relationship | |
| from .database import Base | |
| def _utcnow() -> datetime: | |
| return datetime.now(timezone.utc) | |
| class Document(Base): | |
| __tablename__ = "documents" | |
| id = Column(Integer, primary_key=True, index=True) | |
| filename = Column(String, nullable=False) | |
| file_type = Column(String, nullable=False) # "pdf" or "txt" | |
| status = Column(String, nullable=False, default="processed") # processed | failed | |
| error = Column(Text, nullable=True) | |
| num_chunks = Column(Integer, nullable=False, default=0) | |
| created_at = Column(DateTime, default=_utcnow) | |
| # Owner of the document. NULL = a shared demo/eval document (visible to all). | |
| owner_id = Column( | |
| Integer, ForeignKey("users.id", ondelete="CASCADE"), nullable=True, index=True | |
| ) | |
| owner = relationship("User", back_populates="documents") | |
| chunks = relationship( | |
| "Chunk", back_populates="document", cascade="all, delete-orphan" | |
| ) | |
| questions = relationship( | |
| "QA", back_populates="document", cascade="all, delete-orphan" | |
| ) | |
| class Chunk(Base): | |
| __tablename__ = "chunks" | |
| id = Column(Integer, primary_key=True, index=True) | |
| document_id = Column( | |
| Integer, ForeignKey("documents.id", ondelete="CASCADE"), index=True | |
| ) | |
| chunk_index = Column(Integer, nullable=False) # order within the document | |
| page = Column(Integer, nullable=True) # 1-based page number when known | |
| text = Column(Text, nullable=False) | |
| # Optional dense embedding (float32 bytes) for hybrid retrieval. NULL when | |
| # embeddings are disabled/unavailable at ingest — retrieval then stays lexical. | |
| embedding = Column(LargeBinary, nullable=True) | |
| document = relationship("Document", back_populates="chunks") | |
| class QA(Base): | |
| __tablename__ = "qa_history" | |
| id = Column(Integer, primary_key=True, index=True) | |
| document_id = Column( | |
| Integer, ForeignKey("documents.id", ondelete="CASCADE"), index=True | |
| ) | |
| # The asker — so history is scoped per user even over shared demo documents. | |
| user_id = Column( | |
| Integer, ForeignKey("users.id", ondelete="CASCADE"), nullable=True, index=True | |
| ) | |
| question = Column(Text, nullable=False) | |
| answer = Column(Text, nullable=False) | |
| abstained = Column(Integer, nullable=False, default=0) # 0/1 boolean | |
| citations_json = Column(Text, nullable=False, default="[]") | |
| created_at = Column(DateTime, default=_utcnow) | |
| document = relationship("Document", back_populates="questions") | |
| class User(Base): | |
| __tablename__ = "users" | |
| id = Column(Integer, primary_key=True, index=True) | |
| email = Column(String, nullable=False, unique=True, index=True) | |
| # Nullable: Google-only accounts have no password. | |
| password_hash = Column(String, nullable=True) | |
| created_at = Column(DateTime, default=_utcnow) | |
| documents = relationship( | |
| "Document", back_populates="owner", cascade="all, delete-orphan" | |
| ) | |