mcelwain
/
document-processor
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Create vision-assisted layout candidate scaffold

This commit is contained in:
Sean McElwain 2026-05-30 15:46:12 -05:00
parent 49256cb676
commit ff228b9481
1 changed files with 48 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

63
app/routes/documents.py
View File

@ -2224,35 +2224,68 @@ def document_detail(document_id: str, request: Request, queue: str | None = None
document_for_vision = db.query(Document).filter(Document.document_id == document_id).first() document_for_vision = db.query(Document).filter(Document.document_id == document_id).first()
if document_for_vision: if document_for_vision:
DocumentVisionAnalysisOutput.__table__.create(bind=db.get_bind(), checkfirst=True) DocumentVisionAnalysisOutput.__table__.create(bind=db.get_bind(), checkfirst=True)
source_version = (
db.query(TextVersion)
.filter(TextVersion.document_id == document_for_vision.id)
.filter(TextVersion.layout_json.isnot(None))
.order_by(TextVersion.id.desc())
.first()
)
source_layout = source_version.layout_json if source_version and isinstance(source_version.layout_json, dict) else {"pages": []}
candidate_layout = deepcopy(source_layout)
candidate_layout["vision_assisted"] = True
candidate_layout["vision_assisted_status"] = "placeholder_copied_from_current_layout"
candidate_layout["layout_sync_source"] = "vision_assisted"
candidate_layout["layout_needs_review"] = True
output = DocumentVisionAnalysisOutput( output = DocumentVisionAnalysisOutput(
document_id=document_for_vision.id, document_id=document_for_vision.id,
engine="local_placeholder", engine="local_placeholder",
model_name="none", model_name="none",
prompt_version="vision_scaffold_v1", prompt_version="vision_candidate_v1",
output_type="layout_suggestions", output_type="layout_candidate",
analysis_json={ analysis_json={
"schema_version": "vision_analysis_v1", "schema_version": "vision_analysis_v1",
"status": "placeholder", "status": "candidate_created_from_current_layout",
"document_id": document_id, "document_id": document_id,
"layers": { "source_text_version_id": source_version.id if source_version else None,
"vision_boxes": [], "candidate_kind": "layout_json_copy_placeholder",
"vision_lines": [],
"vision_regions": [
{"bbox": [0, 0, 1200, 700], "label": "VISION TEST REGION"}
],
"vision_fields": [],
"vision_line_items": [],
},
"notes": [ "notes": [
"Vision analysis scaffold created.", "Vision candidate scaffold created.",
"Next step: render page image and populate this JSON from local CV/VLM output.", "Next step: replace copied boxes with local CV/Ollama-derived boxes and merge scoring."
], ],
}, },
created_by="layout_review_run_vision", created_by="layout_review_run_vision",
) )
db.add(output) db.add(output)
db.flush()
next_version_number = (
(db.query(func.max(TextVersion.version_number))
.filter(TextVersion.document_id == document_for_vision.id)
.scalar() or 0) + 1
)
candidate_text = _canonical_layout_text(candidate_layout)
candidate = TextVersion(
document_id=document_for_vision.id,
version_type="vision_assisted_layout",
version_number=next_version_number,
text_content=candidate_text,
layout_json=candidate_layout,
created_by="vision_assisted_layout_candidate",
is_current=True,
rerun_source="vision_assisted",
derived_from_version_id=source_version.id if source_version else None,
)
db.query(TextVersion).filter(TextVersion.document_id == document_for_vision.id).update({"is_current": False})
db.add(candidate)
db.commit() db.commit()
print(f"[vision-analysis] stored placeholder output id={output.id} for {document_id}", flush=True)
print(f"[vision-analysis] stored output id={output.id} and candidate text_version for {document_id}", flush=True)
return RedirectResponse( return RedirectResponse(
url=f"/documents/{document_id}?tab=layout-review&success=vision_analysis_started", url=f"/documents/{document_id}?tab=layout-review&success=vision_analysis_started",