Celes Renata
a72f336ad1
Multi-stage evidence-grounded inference architecture replacing the monolithic 9B model extraction pipeline. CPU-first specialist services handle routine extraction while the 9B vLLM model is preserved for semantic adjudication of ambiguous cases. Key components: - Capability-aware inference gateway (OpenAI-compatible + Ollama) - Endpoint registry with DB migrations and REST API - Sentence-aware document segmenter (property tests) - Deterministic financial parsing with offset integrity - Symbol resolution with ambiguity detection - Specialist service (GLiNER2, dynamic batching, K8s deployment) - Company-specific sentiment (FinBERT, calibration) - Retrieval-based novelty and duplicate detection - Confidence calibration pipeline - Deterministic routing engine (property tests) - 9B adjudication layer with VRAM gating - Stock-specific impact model (features, labels, baseline, trained) - Pipeline orchestrator (state machine, queues, leases, feature flags) - Bounded parallelism (async workers, semaphore, load shedding) - Observability (tracing, metrics, alerts) - Compatibility adapter (v3→v2 golden mapping tests) - Shadow/canary promotion framework - Active learning and fine-tuning pipeline Test results: 1,161 tests pass, ruff lint clean. All 282 spec tasks completed.
76 lines
3.0 KiB
Python
76 lines
3.0 KiB
Python
"""Pydantic models for novelty and duplicate detection."""
|
|
|
|
from __future__ import annotations
|
|
|
|
from pydantic import BaseModel, Field, field_validator
|
|
|
|
|
|
class NearestMatch(BaseModel):
|
|
"""A single nearest-neighbor match from the novelty index."""
|
|
|
|
doc_id: str = Field(description="Document or event identifier of the match")
|
|
similarity_score: float = Field(
|
|
ge=0.0, le=1.0, description="Cosine similarity score (0=unrelated, 1=identical)"
|
|
)
|
|
metadata: dict = Field(default_factory=dict, description="Optional metadata about the match")
|
|
|
|
|
|
class NoveltyResult(BaseModel):
|
|
"""Result of novelty scoring for a document or event.
|
|
|
|
Novelty is scored from 0 (exact duplicate) to 1 (completely novel).
|
|
The formula version tracks which scoring algorithm produced the result.
|
|
"""
|
|
|
|
document_novelty: float = Field(
|
|
ge=0.0, le=1.0, description="Document-level novelty (0=duplicate, 1=novel)"
|
|
)
|
|
event_novelty: float = Field(
|
|
ge=0.0, le=1.0, description="Event-level novelty (0=duplicate event, 1=novel event)"
|
|
)
|
|
combined_novelty: float = Field(
|
|
ge=0.0, le=1.0, description="Combined novelty score used downstream"
|
|
)
|
|
nearest_matches: list[NearestMatch] = Field(
|
|
default_factory=list, description="Nearest matches for explainability"
|
|
)
|
|
formula_version: str = Field(description="Versioned identifier for the novelty formula used")
|
|
is_exact_duplicate: bool = Field(
|
|
default=False, description="Whether an exact content fingerprint match was found"
|
|
)
|
|
is_near_duplicate: bool = Field(
|
|
default=False, description="Whether a near-duplicate fingerprint match was found"
|
|
)
|
|
|
|
@field_validator("nearest_matches")
|
|
@classmethod
|
|
def matches_sorted_descending(cls, v: list[NearestMatch]) -> list[NearestMatch]:
|
|
"""Ensure nearest matches are sorted by similarity descending."""
|
|
return sorted(v, key=lambda m: m.similarity_score, reverse=True)
|
|
|
|
|
|
class FingerprintRecord(BaseModel):
|
|
"""Stored fingerprint for a document."""
|
|
|
|
doc_id: str = Field(description="Document identifier")
|
|
exact_fingerprint: str = Field(description="SHA-256 of normalized text")
|
|
simhash: int = Field(description="SimHash value for near-duplicate detection")
|
|
metadata: dict = Field(default_factory=dict, description="Additional metadata")
|
|
|
|
|
|
class EmbeddingRecord(BaseModel):
|
|
"""Stored embedding vector for a document or event."""
|
|
|
|
doc_id: str = Field(description="Document or event identifier")
|
|
embedding: list[float] = Field(description="Dense embedding vector")
|
|
record_type: str = Field(description="Type: 'document' or 'company_event'")
|
|
metadata: dict = Field(default_factory=dict, description="Additional metadata")
|
|
|
|
@field_validator("record_type")
|
|
@classmethod
|
|
def valid_record_type(cls, v: str) -> str:
|
|
valid = {"document", "company_event"}
|
|
if v not in valid:
|
|
raise ValueError(f"record_type must be one of {valid}, got '{v}'")
|
|
return v
|