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.
81 lines
2.5 KiB
Python
81 lines
2.5 KiB
Python
"""Data models for symbol resolution results."""
|
|
|
|
from __future__ import annotations
|
|
|
|
from enum import Enum
|
|
|
|
from pydantic import BaseModel, Field
|
|
|
|
|
|
class MatchType(str, Enum):
|
|
"""How a resolution candidate was matched."""
|
|
|
|
exact_ticker = "exact_ticker"
|
|
exact_name = "exact_name"
|
|
alias = "alias"
|
|
fuzzy = "fuzzy"
|
|
|
|
|
|
class MentionType(str, Enum):
|
|
"""Whether a mention is explicitly stated or inferred from context."""
|
|
|
|
explicit = "explicit"
|
|
inferred = "inferred"
|
|
|
|
|
|
class UnresolvedReason(str, Enum):
|
|
"""Why a mention could not be resolved."""
|
|
|
|
not_in_registry = "not_in_registry"
|
|
ambiguous = "ambiguous"
|
|
context_needed = "context_needed"
|
|
|
|
|
|
class ResolutionCandidate(BaseModel):
|
|
"""A single candidate match from the symbol registry.
|
|
|
|
Candidates are ranked by confidence, with match_type indicating how
|
|
the match was derived.
|
|
"""
|
|
|
|
company_id: str = Field(description="UUID of the matched company")
|
|
ticker: str = Field(description="Ticker symbol of the matched company")
|
|
name: str = Field(description="Legal or display name of the matched company")
|
|
confidence: float = Field(ge=0.0, le=1.0, description="Match confidence score")
|
|
match_type: MatchType = Field(description="How the match was derived")
|
|
|
|
|
|
class ResolutionResult(BaseModel):
|
|
"""Full resolution output for a single mention.
|
|
|
|
Contains ranked candidates, ambiguity margin, and mention classification.
|
|
"""
|
|
|
|
candidates: list[ResolutionCandidate] = Field(default_factory=list)
|
|
ambiguity_margin: float = Field(
|
|
default=1.0,
|
|
ge=0.0,
|
|
le=1.0,
|
|
description="Difference between top-2 candidate confidences. 1.0 = unambiguous single match, 0.0 = tied.",
|
|
)
|
|
is_ambiguous: bool = Field(
|
|
default=False,
|
|
description="True when top candidates are too close to distinguish without context.",
|
|
)
|
|
mention_type: MentionType = Field(
|
|
default=MentionType.explicit,
|
|
description="Whether the mention is explicit text or inferred exposure.",
|
|
)
|
|
|
|
|
|
class UnresolvedMention(BaseModel):
|
|
"""A mention that could not be resolved to any company in the registry.
|
|
|
|
Preserved as-is rather than having a ticker invented for it.
|
|
"""
|
|
|
|
literal_text: str = Field(description="Original text as it appeared in the document")
|
|
start_char: int = Field(ge=0, description="Start character offset in source document")
|
|
end_char: int = Field(gt=0, description="End character offset (exclusive)")
|
|
reason: UnresolvedReason = Field(description="Why resolution failed")
|