Added markdown builder (codex)

README.md

@@ -417,6 +417,8 @@ Telegram ботом, которому отправили видео).
 9. **Собрать финальный Markdown файл**
     - Для сборки Markdown файла используются элементы, полученные на предыдщем
       шаге, и хранимые в файле `structure_refined.json`
+    - Сборка выполняется локально, без использования ИИ. Ссылки на изображения
+      разрешаются по `image.event_id` через `events.json`.
     - В итоге создаётся файл `output.md`, который может включать в себя ссылки
       на изображения из директории `images/` (относительно директории
       промежуточных данных)

main.py

@@ -16,6 +16,7 @@ from asr_filter import AsrFilter, AsrFilterResult
 from asr_eventizer import AsrEventizer
 from structure_builder import Structure, StructureBuilder
 from structure_refiner import StructureRefiner
+from markdown_builder import MarkdownBuilder
 from windowizer import Windowizer
 
 from agent import Agent
@@ -272,10 +273,23 @@ def on_structure_refiner(current_step: Step, input_data: dict | None) -> tuple[S
     result = refiner.refine(Structure(**input_data))
     return (Step.MARKDOWN_BUILDER, result.model_dump(mode="json"))
 
+def on_markdown_builder(current_step: Step, input_data: dict | None) -> tuple[Step | None, str | None]:
+    if os.path.isfile(WORKFLOW_DATA[current_step][0]):
+        logging.info("Skipping Markdown builder")
+        return (None, None)
+    if input_data is None:
+        logging.error("Can't build Markdown without input_data")
+        return (None, None)
+    with open("events.json", "rb") as f:
+        timeline = Timeline(**json.load(f))
+    logging.info("Building Markdown...")
+    builder = MarkdownBuilder(timeline)
+    return (None, builder.build(Structure(**input_data)))
+
 #
 # Main
 #
-WORKFLOW_DATA: dict[Step, tuple[str, Callable[[Step, dict | None], tuple[Step | None, dict | None]] | None]] = {
+WORKFLOW_DATA: dict[Step, tuple[str, Callable[[Step, dict | None], tuple[Step | None, dict | str | None]] | None]] = {
     Step.MEDIA_SEPARATION: ("audio.mp3", on_media_separation),
     Step.VOICE_RECOGNITION: ("asr_raw.json", on_voice_recognition),
     Step.ASR_FILTER: ("asr.json", on_asr_filter),
@@ -284,7 +298,7 @@ WORKFLOW_DATA: dict[Step, tuple[str, Callable[[Step, dict | None], tuple[Step |
     Step.REFERENCE_RESOLVER: ("events.json", on_reference_resolver),
     Step.STRUCTURE_BUILDER: ("structure.json", on_structure_builder),
     Step.STRUCTURE_REFINER: ("structure_refined.json", on_structure_refiner),
-    Step.MARKDOWN_BUILDER: ("output.md", None)
+    Step.MARKDOWN_BUILDER: ("output.md", on_markdown_builder)
 }
 """Information about workflow.
 
@@ -314,8 +328,11 @@ def main() -> None:
         logging.info(f"Executing step {step_to_do}")
         step_to_do, ret = func(step_to_do, intermediate_result)
         if ret is not None:
+            with open(output_file_path, "w", encoding="utf-8") as f:
+                if isinstance(ret, str):
+                    f.write(ret)
+                else:
                     intermediate_result = ret
-            with open(output_file_path, "w") as f:
                     json.dump(ret, f, ensure_ascii=False, indent=4)
             logging.info(f"Intermediate results are saved {output_file_path}")
     # final report

markdown_builder.py

@@ -0,0 +1,59 @@
+from structure_builder import (
+    DefinitionElement,
+    HeadingElement,
+    ImageElement,
+    ImportantElement,
+    OrderedListElement,
+    ParagraphElement,
+    Structure,
+    StructureElement,
+    UnorderedListElement,
+)
+from utils import Event, Timeline
+
+
+class MarkdownBuilder:
+    """Render a document structure to Markdown without using AI."""
+
+    def __init__(self, timeline: Timeline) -> None:
+        self._events: dict[int, Event] = {}
+        for event in timeline.events:
+            if event.id in self._events:
+                raise ValueError(f"Duplicate event ID: {event.id}")
+            self._events[event.id] = event
+
+    def _render_image(self, element: ImageElement) -> str:
+        event = self._events.get(element.event_id)
+        if event is None or event.type != "vis" or not event.payload:
+            raise ValueError(
+                f"Image element references invalid visual event: {element.event_id}"
+            )
+        alt = event.text.replace("[", "").replace("]", "").replace("\n", " ")
+        return f"![{alt}](<{event.payload}>)"
+
+    def _render_element(self, element: StructureElement) -> str:
+        if isinstance(element, HeadingElement):
+            return f"{'#' * element.level} {element.text}"
+        if isinstance(element, ParagraphElement):
+            return element.text
+        if isinstance(element, UnorderedListElement):
+            return "\n".join(f"- {item}" for item in element.items)
+        if isinstance(element, OrderedListElement):
+            return "\n".join(
+                f"{index}. {item}"
+                for index, item in enumerate(element.items, start=1)
+            )
+        if isinstance(element, DefinitionElement):
+            return f"**{element.term}.** {element.text}"
+        if isinstance(element, ImportantElement):
+            text = f"**Важно:** {element.text}"
+            return "\n".join(f"> {line}" for line in text.splitlines())
+        if isinstance(element, ImageElement):
+            return self._render_image(element)
+        raise TypeError(f"Unsupported structure element: {type(element).__name__}")
+
+    def build(self, structure: Structure) -> str:
+        blocks = [self._render_element(element) for element in structure.elements]
+        if not blocks:
+            return ""
+        return "\n\n".join(blocks) + "\n"