fix(similarity): address code review - column-only queries, configurable batch size, WCAG touch targets

- Use column-only query in embeddings overview to reduce memory for 100K+ files
- Add embedding_backfill_batch_size config setting (default 50)
- Fix WCAG touch target on backfill button (min-height/min-width 44px)
- Add inline comment explaining 3 chars/token truncation estimate
- Import settings in compute_embedding task for configurable batch size

Co-authored-by: christianlouis <361235+christianlouis@users.noreply.github.com>

app/api/similarity.py

@@ -257,7 +257,17 @@ def get_embeddings_overview(
     }
     ```
     """
-    all_files = db.query(FileRecord).order_by(FileRecord.id.desc()).all()
+    # Use column-only query to avoid loading full ORM objects into memory
+    all_files = (
+        db.query(
+            FileRecord.id,
+            FileRecord.original_filename,
+            FileRecord.ocr_text,
+            FileRecord.embedding,
+        )
+        .order_by(FileRecord.id.desc())
+        .all()
+    )
 
     files_info = []
     total_with_ocr = 0

app/config.py

@@ -339,6 +339,13 @@ class Settings(BaseSettings):
             "Set this below the model's context window (e.g. 8000 for an 8192-token model)."
         ),
     )
+    embedding_backfill_batch_size: int = Field(
+        default=50,
+        description=(
+            "Maximum number of files to queue for embedding computation per "
+            "backfill run.  Keeps the worker and embedding API load bounded."
+        ),
+    )
 
     # Text quality check - AI-based assessment of embedded PDF text
     enable_text_quality_check: bool = Field(

app/tasks/compute_embedding.py

@@ -9,6 +9,7 @@ import logging
 from datetime import datetime, timezone
 
 from app.celery_app import celery
+from app.config import settings
 from app.database import SessionLocal
 from app.models import FileRecord
 from app.tasks.retry_config import BaseTaskWithRetry
@@ -145,7 +146,7 @@ def backfill_missing_embeddings(self) -> dict:
     Returns:
         A dict with the number of tasks ``queued``.
     """
-    batch_size = 50  # max files to queue per run
+    batch_size = settings.embedding_backfill_batch_size
     task_id = self.request.id
     logger.info("[%s] Backfill: scanning for files missing embeddings (batch_size=%d)", task_id, batch_size)
 

app/utils/similarity.py

@@ -59,8 +59,8 @@ def generate_embedding(text: str, model: str | None = None) -> list[float]:
         model = settings.embedding_model
 
     # Truncate to stay within the model's context window.
-    # Use a conservative estimate of ~3 characters per token so that the
+    # Conservative 3 chars/token estimate (actual ratio varies by language;
-    # resulting text fits comfortably within ``embedding_max_tokens``.
+    # English averages ~4 chars/token but 3 gives a safety margin).
     max_chars = settings.embedding_max_tokens * 3
     if len(text) > max_chars:
         logger.debug(

frontend/templates/similarity_dashboard.html

@@ -133,7 +133,7 @@
     <strong>{{ files_missing_embedding }}</strong> file(s) have OCR text but no embedding yet.
     The background task will compute them automatically every 5 minutes, or you can
     <button onclick="triggerBackfill()" id="backfill-btn"
-            style="background: #d97706; color: white; border: none; padding: 0.2rem 0.6rem; border-radius: 0.25rem; cursor: pointer; font-size: 0.8rem;"
+            style="background: #d97706; color: white; border: none; padding: 0.2rem 0.6rem; border-radius: 0.25rem; cursor: pointer; font-size: 0.8rem; min-height: 44px; min-width: 44px;"
             aria-label="Trigger embedding computation for all files missing embeddings">
       trigger it now
     </button>.