src/interfaces/composition.py - main - artsadert/sber_hackathon

src/interfaces/composition.py

41 строка2 KB

Gleb

Integrate MinIO for artifact storage and enhance report generation. Updated .env.example for MinIO configuration, added MinIO service to docker-compose, and modified the pipeline to include source references and causal graphs in reports. Updated dependencies in pyproject.toml and adjusted frontend components for new report structure.

06 июн 2026, 16:57

06 июн 2026, 16:57e8f1a58

from pathlib import Path

from application.pipeline import AnalyzePipeline
from domain.models import Report
from domain.ports import TokenCounter
from infrastructure.chunk.chunker import TokenChunker
from infrastructure.ingest.format_detector import detect
from infrastructure.ingest.loader import DocumentLoader
from infrastructure.llm.lm import LLMRunner
from infrastructure.llm.modules import (
    DspyEventExtractor,
    DspyReportWriter,
    DspyRootCauseAnalyzer,
)
from infrastructure.llm.tokenizer import get_token_counter
from infrastructure.preprocess.masker import ConsistentMasker
from infrastructure.preprocess.preprocessor import LogPreprocessor, PassthroughPreprocessor


def build_pipeline(
    file_type: str,
    runner: LLMRunner,
    counter: TokenCounter | None = None,
) -> AnalyzePipeline:
    preprocessor = PassthroughPreprocessor() if file_type == "table" else LogPreprocessor()
    return AnalyzePipeline(
        masker=ConsistentMasker(),
        preprocessor=preprocessor,
        chunker=TokenChunker(counter or get_token_counter()),
        extractor=DspyEventExtractor(runner),
        analyzer=DspyRootCauseAnalyzer(runner),
        reporter=DspyReportWriter(runner),
    )


async def analyze_file(path: str, source_name: str | None = None) -> Report:
    file_type = detect(path)
    source_text = None if file_type == "table" and Path(path).suffix.lower() in {".xlsx", ".xls"} else Path(path).read_text(encoding="utf-8")
    records = DocumentLoader().load(path)
    pipeline = build_pipeline(file_type, LLMRunner())
    return await pipeline.run(records, source_name=source_name or Path(path).name, source_text=source_text)