Upload 3 files

Dockerfile

@@ -1,21 +1,13 @@
-# ─────────────────────────────────────────────────────────
-# ValuationAI OCR — Docker Image
-# Model: rasmodev/Handwriting_trocr_model
-# ─────────────────────────────────────────────────────────
 FROM python:3.11-slim
 
 RUN apt-get update && apt-get install -y --no-install-recommends \
-        libgl1 \
-        libglib2.0-0 \
-        libsm6 \
-        libxext6 \
-        libxrender-dev \
-        curl \
+        libgl1 libglib2.0-0 libsm6 libxext6 libxrender-dev curl \
     && rm -rf /var/lib/apt/lists/*
 
 RUN useradd -m -u 1000 appuser
 WORKDIR /app
 
+# CPU PyTorch
 RUN pip install --no-cache-dir \
     torch==2.2.2+cpu torchvision==0.17.2+cpu \
     --extra-index-url https://download.pytorch.org/whl/cpu
@@ -25,17 +17,18 @@ RUN pip install --no-cache-dir -r requirements.txt
 
 COPY app.py .
 
-ENV HF_HOME=/app/.cache/huggingface
-RUN mkdir -p /app/.cache/huggingface && chown -R appuser:appuser /app
+# Cache directories writable by non-root user
+ENV HF_HOME=/tmp/huggingface
+ENV TRANSFORMERS_CACHE=/tmp/huggingface/transformers
+ENV HF_DATASETS_CACHE=/tmp/huggingface/datasets
+RUN mkdir -p /tmp/huggingface && chmod -R 777 /tmp/huggingface
 
 USER appuser
 EXPOSE 7860
 
-HEALTHCHECK --interval=30s --timeout=10s --start-period=90s \
-    CMD curl -f http://localhost:7860/_stcore/health || exit 1
-
 CMD ["streamlit", "run", "app.py", \
      "--server.port=7860", \
      "--server.address=0.0.0.0", \
      "--server.headless=true", \
-     "--browser.gatherUsageStats=false"]
+     "--server.fileWatcherType=none", \
+     "--browser.gatherUsageStats=false"]

app.py

@@ -1,6 +1,9 @@
 """
 ValuationAI — Nairobi Valuation Sheet OCR
 Model: rasmodev/Handwriting_trocr_model
+
+Label format from training:
+PLOT: LR 209/617 | LOC: STATE HOUSE AVENUE | AREA: 0.06 | AMT: 52000000 | DATE: 2008-06-17 | VOS: 3872
 """
 import io, time, logging
 import streamlit as st
@@ -17,81 +20,41 @@ logging.basicConfig(level=logging.INFO)
 
 st.markdown("""
 <style>
-@import url('https://fonts.googleapis.com/css2?family=Cormorant+Garamond:wght@400;600;700&family=Inter:wght@300;400;500;600&display=swap');
+@import url('https://fonts.googleapis.com/css2?family=Cormorant+Garamond:ital,wght@0,600;0,700;1,600&family=Inter:wght@300;400;500;600&display=swap');
 
 html, body, [class*="css"], .stApp {
     font-family: 'Inter', sans-serif;
     background: #F8F7F4;
     color: #1A1A2E;
 }
-
 .block-container {
     padding: 3rem 4rem !important;
     max-width: 1100px !important;
 }
-
 #MainMenu, footer, header { visibility: hidden; }
 
-/* ── Top bar ── */
+/* Top bar */
 .topbar {
-    display: flex;
-    align-items: flex-end;
+    display: flex; align-items: flex-end;
     justify-content: space-between;
-    padding-bottom: 2rem;
-    margin-bottom: 3rem;
+    padding-bottom: 2rem; margin-bottom: 3rem;
     border-bottom: 2px solid #1A1A2E;
 }
-.logo {
-    font-family: 'Cormorant Garamond', serif;
-    font-size: 1.8rem;
-    font-weight: 700;
-    color: #1A1A2E;
-    letter-spacing: -0.02em;
-    line-height: 1;
-}
+.logo { font-family: 'Cormorant Garamond', serif; font-size: 1.8rem; font-weight: 700; color: #1A1A2E; letter-spacing: -0.02em; line-height: 1; }
 .logo span { color: #2563EB; }
-.logo-sub {
-    font-size: 0.68rem;
-    font-weight: 500;
-    letter-spacing: 0.15em;
-    text-transform: uppercase;
-    color: #9CA3AF;
-    margin-top: 0.3rem;
-}
-.model-ref {
-    font-size: 0.7rem;
-    color: #9CA3AF;
-    font-weight: 400;
-    letter-spacing: 0.04em;
-    text-align: right;
-}
+.logo-sub { font-size: 0.68rem; font-weight: 500; letter-spacing: 0.15em; text-transform: uppercase; color: #9CA3AF; margin-top: 0.3rem; }
+.model-ref { font-size: 0.7rem; color: #9CA3AF; font-weight: 400; letter-spacing: 0.04em; text-align: right; }
 .model-ref strong { color: #2563EB; font-weight: 600; }
 
-/* ── Hero headline ── */
-.headline {
-    font-family: 'Cormorant Garamond', serif;
-    font-size: 3.4rem;
-    font-weight: 700;
-    line-height: 1.08;
-    letter-spacing: -0.03em;
-    color: #1A1A2E;
-    margin-bottom: 1rem;
-    max-width: 700px;
-}
-.headline em {
-    font-style: italic;
-    color: #2563EB;
-}
-.subline {
-    font-size: 0.95rem;
-    font-weight: 300;
-    color: #6B7280;
-    line-height: 1.7;
-    max-width: 500px;
-    margin-bottom: 3rem;
-}
+/* Headline */
+.headline { font-family: 'Cormorant Garamond', serif; font-size: 3.4rem; font-weight: 700; line-height: 1.08; letter-spacing: -0.03em; color: #1A1A2E; margin-bottom: 1rem; max-width: 700px; }
+.headline em { font-style: italic; color: #2563EB; }
+.subline { font-size: 0.95rem; font-weight: 300; color: #6B7280; line-height: 1.7; max-width: 500px; margin-bottom: 3rem; }
+
+/* Step label */
+.step { font-size: 0.65rem; font-weight: 700; letter-spacing: 0.18em; text-transform: uppercase; color: #2563EB; margin-bottom: 0.5rem; }
 
-/* ── Upload area ── */
+/* File uploader */
 [data-testid="stFileUploader"] section {
     background: #fff !important;
     border: 2px dashed #D1D5DB !important;
@@ -104,166 +67,56 @@ html, body, [class*="css"], .stApp {
     border-color: #2563EB !important;
     box-shadow: 0 0 0 4px rgba(37,99,235,0.06) !important;
 }
-[data-testid="stFileUploader"] label {
-    color: #6B7280 !important;
-    font-size: 0.9rem !important;
-}
-[data-testid="stFileUploadDropzone"] p {
-    color: #6B7280 !important;
-}
+[data-testid="stFileUploader"] label { color: #6B7280 !important; font-size: 0.9rem !important; }
+
+/* File chip */
+.fchip { display: inline-flex; align-items: center; gap: 5px; background: #EFF6FF; border: 1px solid #BFDBFE; color: #1D4ED8; padding: 0.25rem 0.7rem; border-radius: 6px; font-size: 0.73rem; font-weight: 500; margin: 2px; }
 
-/* ── Run button ── */
+/* Button */
 .stButton > button {
-    background: #1A1A2E !important;
-    color: #fff !important;
-    border: none !important;
-    border-radius: 8px !important;
-    padding: 0.85rem 2.5rem !important;
-    font-family: 'Inter', sans-serif !important;
-    font-size: 0.88rem !important;
-    font-weight: 600 !important;
-    letter-spacing: 0.04em !important;
-    text-transform: uppercase !important;
-    transition: all 0.2s !important;
-    box-shadow: 0 2px 8px rgba(26,26,46,0.2) !important;
-    width: 100% !important;
-}
-.stButton > button:hover {
-    background: #2563EB !important;
-    box-shadow: 0 4px 16px rgba(37,99,235,0.3) !important;
-    transform: translateY(-1px) !important;
-}
-.stButton > button:disabled {
-    background: #E5E7EB !important;
-    color: #9CA3AF !important;
-    box-shadow: none !important;
-    transform: none !important;
-    cursor: not-allowed !important;
+    background: #1A1A2E !important; color: #fff !important; border: none !important;
+    border-radius: 8px !important; padding: 0.85rem 2.5rem !important;
+    font-family: 'Inter', sans-serif !important; font-size: 0.88rem !important;
+    font-weight: 600 !important; letter-spacing: 0.04em !important;
+    text-transform: uppercase !important; transition: all 0.2s !important;
+    box-shadow: 0 2px 8px rgba(26,26,46,0.2) !important; width: 100% !important;
 }
+.stButton > button:hover { background: #2563EB !important; box-shadow: 0 4px 16px rgba(37,99,235,0.3) !important; transform: translateY(-1px) !important; }
+.stButton > button:disabled { background: #E5E7EB !important; color: #9CA3AF !important; box-shadow: none !important; transform: none !important; }
 
-/* ── Progress ── */
-.stProgress > div > div > div {
-    background: #2563EB !important;
-    border-radius: 4px !important;
-}
-.stProgress > div > div {
-    background: #E5E7EB !important;
-    border-radius: 4px !important;
-    height: 4px !important;
-}
+/* Progress */
+.stProgress > div > div > div { background: #2563EB !important; border-radius: 4px !important; }
+.stProgress > div > div { background: #E5E7EB !important; border-radius: 4px !important; height: 4px !important; }
 
-/* ── Stats strip ── */
-.stats-strip {
-    display: flex;
-    gap: 0;
-    background: #1A1A2E;
-    border-radius: 12px;
-    overflow: hidden;
-    margin: 2.5rem 0 2rem;
-}
-.stat-item {
-    flex: 1;
-    padding: 1.6rem 2rem;
-    border-right: 1px solid rgba(255,255,255,0.08);
-    text-align: left;
-}
+/* Stats */
+.stats-strip { display: flex; background: #1A1A2E; border-radius: 12px; overflow: hidden; margin: 2.5rem 0 2rem; }
+.stat-item { flex: 1; padding: 1.6rem 2rem; border-right: 1px solid rgba(255,255,255,0.08); }
 .stat-item:last-child { border-right: none; }
-.stat-n {
-    font-family: 'Cormorant Garamond', serif;
-    font-size: 2.6rem;
-    font-weight: 700;
-    color: #fff;
-    line-height: 1;
-    margin-bottom: 0.3rem;
-}
-.stat-l {
-    font-size: 0.68rem;
-    font-weight: 500;
-    letter-spacing: 0.12em;
-    text-transform: uppercase;
-    color: #6B7280;
-}
+.stat-n { font-family: 'Cormorant Garamond', serif; font-size: 2.6rem; font-weight: 700; color: #fff; line-height: 1; margin-bottom: 0.3rem; }
+.stat-l { font-size: 0.68rem; font-weight: 500; letter-spacing: 0.12em; text-transform: uppercase; color: #6B7280; }
 
-/* ── Section heading ── */
-.section-head {
-    display: flex;
-    align-items: center;
-    justify-content: space-between;
-    margin-bottom: 1rem;
-    padding-bottom: 0.75rem;
-    border-bottom: 1px solid #E5E7EB;
-}
-.section-title {
-    font-family: 'Cormorant Garamond', serif;
-    font-size: 1.5rem;
-    font-weight: 600;
-    color: #1A1A2E;
-    letter-spacing: -0.01em;
-}
+/* Section head */
+.section-head { display: flex; align-items: center; justify-content: space-between; margin-bottom: 1rem; padding-bottom: 0.75rem; border-bottom: 1px solid #E5E7EB; }
+.section-title { font-family: 'Cormorant Garamond', serif; font-size: 1.5rem; font-weight: 600; color: #1A1A2E; letter-spacing: -0.01em; }
 
-/* ── Download button ── */
+/* Download button */
 div[data-testid="stDownloadButton"] > button {
-    background: #fff !important;
-    border: 1.5px solid #1A1A2E !important;
-    color: #1A1A2E !important;
-    border-radius: 8px !important;
-    padding: 0.6rem 1.4rem !important;
-    font-family: 'Inter', sans-serif !important;
-    font-weight: 600 !important;
-    font-size: 0.82rem !important;
-    letter-spacing: 0.04em !important;
-    text-transform: uppercase !important;
-    transition: all 0.2s !important;
-    box-shadow: none !important;
-    width: auto !important;
-}
-div[data-testid="stDownloadButton"] > button:hover {
-    background: #1A1A2E !important;
-    color: #fff !important;
+    background: #fff !important; border: 1.5px solid #1A1A2E !important; color: #1A1A2E !important;
+    border-radius: 8px !important; padding: 0.6rem 1.4rem !important;
+    font-family: 'Inter', sans-serif !important; font-weight: 600 !important;
+    font-size: 0.82rem !important; letter-spacing: 0.04em !important;
+    text-transform: uppercase !important; transition: all 0.2s !important; width: auto !important;
 }
+div[data-testid="stDownloadButton"] > button:hover { background: #1A1A2E !important; color: #fff !important; }
 
-/* ── Dataframe ── */
-[data-testid="stDataFrame"] {
-    border-radius: 10px !important;
-    border: 1px solid #E5E7EB !important;
-    overflow: hidden !important;
-    box-shadow: 0 1px 4px rgba(0,0,0,0.05) !important;
-}
-
-/* ── File chip ── */
-.fchip {
-    display: inline-flex; align-items: center; gap: 5px;
-    background: #EFF6FF;
-    border: 1px solid #BFDBFE;
-    color: #1D4ED8;
-    padding: 0.25rem 0.7rem;
-    border-radius: 6px;
-    font-size: 0.73rem;
-    font-weight: 500;
-    margin: 2px;
-}
-
-/* ── Divider ── */
-.rule { height:1px; background:#E5E7EB; margin: 2.5rem 0; }
-
-/* ── Step tag ── */
-.step {
-    font-size: 0.65rem; font-weight: 700;
-    letter-spacing: 0.18em; text-transform: uppercase;
-    color: #2563EB; margin-bottom: 0.5rem;
-}
-
-/* ── Success alert ── */
-[data-testid="stAlert"][data-baseweb="notification"] {
-    border-radius: 10px !important;
-    border-left: 3px solid #2563EB !important;
-}
+/* Dataframe */
+[data-testid="stDataFrame"] { border-radius: 10px !important; border: 1px solid #E5E7EB !important; overflow: hidden !important; box-shadow: 0 1px 4px rgba(0,0,0,0.05) !important; }
 </style>
 """, unsafe_allow_html=True)
 
 
 # ═══════════════════════════════════════════════════════════
-# MODEL  (cached — loads once)
+# MODEL LOADING
 # ═══════════════════════════════════════════════════════════
 @st.cache_resource(show_spinner="Loading recognition model…")
 def load_model():
@@ -278,67 +131,116 @@ def load_model():
 
 
 # ═══════════════════════════════════════════════════════════
-# HELPERS
+# OCR — matches training output format exactly
+# Training label format:
+# PLOT: LR 209/617 | LOC: STATE HOUSE AVENUE | AREA: 0.06 | AMT: 52000000 | DATE: 2008-06-17 | VOS: 3872
 # ═══════════════════════════════════════════════════════════
-def ocr(img: Image.Image) -> str:
+def ocr_page(img: Image.Image) -> str:
+    """Run the fine-tuned model on one page image."""
     import torch
     processor, model, device = load_model()
-    pv = processor(images=img.convert("RGB"), return_tensors="pt").pixel_values.to(device)
+    pv = processor(
+        images=img.convert("RGB"),
+        return_tensors="pt"
+    ).pixel_values.to(device)
     with torch.no_grad():
-        gen = model.generate(pv, max_new_tokens=64)
+        gen = model.generate(pv, max_new_tokens=128)
     return processor.batch_decode(gen, skip_special_tokens=True)[0].strip()
 
-def pdf_pages(file_bytes: bytes) -> list:
+
+def parse_label(raw_text: str, filename: str) -> dict:
+    """
+    Parse the pipe-delimited label that the model was trained to output.
+    Format: PLOT: ... | LOC: ... | AREA: ... | AMT: ... | DATE: ... | VOS: ...
+    """
+    record = {
+        "File":         filename,
+        "Plot Number":  "",
+        "Location":     "",
+        "Area":         "",
+        "Amount (KES)": None,
+        "Date":         "",
+        "VOS":          "",
+        "Raw Output":   raw_text,
+    }
+
+    # Split on pipe delimiter
+    parts = raw_text.split("|")
+    for part in parts:
+        part = part.strip()
+        if ":" not in part:
+            continue
+        key, _, val = part.partition(":")
+        key = key.strip().upper()
+        val = val.strip()
+
+        if key == "PLOT":
+            record["Plot Number"] = val
+        elif key == "LOC":
+            record["Location"] = val
+        elif key == "AREA":
+            record["Area"] = val
+        elif key == "AMT":
+            # Remove commas and convert to int
+            try:
+                record["Amount (KES)"] = int(val.replace(",", "").replace(" ", ""))
+            except ValueError:
+                record["Amount (KES)"] = val
+        elif key == "DATE":
+            record["Date"] = val
+        elif key == "VOS":
+            record["VOS"] = val
+
+    return record
+
+
+def pdf_to_images(file_bytes: bytes) -> list:
+    """Convert all pages of a PDF to PIL images at 200 DPI."""
     import fitz
     doc  = fitz.open(stream=file_bytes, filetype="pdf")
     mat  = fitz.Matrix(200/72, 200/72)
-    out  = []
+    imgs = []
     for i in range(len(doc)):
         pix = doc[i].get_pixmap(matrix=mat, alpha=False)
-        out.append(Image.open(io.BytesIO(pix.tobytes("png"))).convert("RGB"))
+        imgs.append(Image.open(io.BytesIO(pix.tobytes("png"))).convert("RGB"))
     doc.close()
-    return out
-
-def extract(text: str, filename: str) -> dict:
-    import re
-    def g(label):
-        m = re.search(label + r'[:\s]+([^\|\n]{1,80})', text, re.IGNORECASE)
-        return m.group(1).strip() if m else ""
-    amounts = re.findall(r'AMT:\s*([\d,]+)', text)
-    date_m  = re.search(r'DATE:\s*(\d{4}-\d{2}-\d{2})', text)
-    return {
-        "File":         filename,
-        "Plot Number":  g("PLOT"),
-        "Location":     g("LOC"),
-        "Area":         g("AREA"),
-        "Amount (KES)": int(amounts[0].replace(",","")) if amounts else None,
-        "Date":         date_m.group(1) if date_m else None,
-        "VOS":          g("VOS"),
-    }
+    return imgs
+
 
 def make_excel(records: list) -> bytes:
+    """Export records to a formatted Excel workbook."""
     from openpyxl import load_workbook
     from openpyxl.styles import Font, PatternFill, Alignment
     from openpyxl.utils import get_column_letter
+
+    # Drop raw output from Excel — it's only for debugging
+    clean = [{k: v for k, v in r.items() if k != "Raw Output"} for r in records]
+
     buf = io.BytesIO()
-    pd.DataFrame(records).to_excel(buf, index=False, sheet_name="Valuation Data")
+    pd.DataFrame(clean).to_excel(buf, index=False, sheet_name="Valuation Data")
     buf.seek(0)
+
     wb  = load_workbook(buf)
     ws  = wb.active
-    hdr_fill = PatternFill("solid", start_color="1A1A2E")
+    hdr = PatternFill("solid", start_color="1A1A2E")
+
     for ci, cell in enumerate(ws[1], 1):
         cell.font      = Font(name="Calibri", bold=True, color="FFFFFF", size=11)
-        cell.fill      = hdr_fill
+        cell.fill      = hdr
         cell.alignment = Alignment(horizontal="center", vertical="center")
-        ws.column_dimensions[get_column_letter(ci)].width = 24
+        ws.column_dimensions[get_column_letter(ci)].width = 26
+
     ws.row_dimensions[1].height = 30
+
     for row in ws.iter_rows(min_row=2):
         for cell in row:
             cell.alignment = Alignment(vertical="center", wrap_text=True)
             if cell.row % 2 == 0:
                 cell.fill = PatternFill("solid", start_color="F0F4FF")
+
     ws.freeze_panes = "A2"
-    out = io.BytesIO(); wb.save(out)
+    out = io.BytesIO()
+    wb.save(out)
     return out.getvalue()
 
 
@@ -351,7 +253,7 @@ for k, v in [("records",[]),("excel",None),("done",False),("errors",[])]:
 
 
 # ═══════════════════════════════════════════════════════════
-# TOP BAR
+# UI — TOP BAR
 # ═══════════════════════════════════════════════════════════
 st.markdown("""
 <div class="topbar">
@@ -368,7 +270,7 @@ st.markdown("""
 
 
 # ═══════════════════════════════════════════════════════════
-# HEADLINE
+# UI — HEADLINE
 # ═══════════════════════════════════════════════════════════
 st.markdown("""
 <div class="headline">
@@ -383,7 +285,7 @@ st.markdown("""
 
 
 # ═══════════════════════════════════════════════════════════
-# UPLOAD
+# UI — UPLOAD
 # ═══════════════════════════════════════════════════════════
 st.markdown('<div class="step">Step 1 — Upload Documents</div>', unsafe_allow_html=True)
 
@@ -424,30 +326,41 @@ if run and uploaded:
     t0     = time.time()
 
     for fi, uf in enumerate(uploaded):
-        fname      = uf.name
-        raw        = uf.read()
+        fname = uf.name
+        raw   = uf.read()
         bar.progress(fi / len(uploaded), text=f"Reading {fname}…")
 
         try:
             ext = fname.lower().rsplit(".", 1)[-1]
-            imgs = pdf_pages(raw) if ext == "pdf" \
-                   else [Image.open(io.BytesIO(raw)).convert("RGB")]
 
+            # Get page images
+            if ext == "pdf":
+                imgs = pdf_to_images(raw)
+            else:
+                imgs = [Image.open(io.BytesIO(raw)).convert("RGB")]
+
+            # Run OCR on each page
             for pi, img in enumerate(imgs, 1):
-                status.caption(f"Processing **{fname}** — page {pi} of {len(imgs)}")
-                text = ocr(img)
-                st.session_state.records.append(extract(text, fname))
+                status.caption(
+                    f"Running recognition on **{fname}** — "
+                    f"page {pi} of {len(imgs)}"
+                )
+                raw_text = ocr_page(img)
+                record   = parse_label(raw_text, fname)
+                st.session_state.records.append(record)
 
         except Exception as e:
             st.session_state.errors.append(f"{fname}: {e}")
 
-        bar.progress((fi+1) / len(uploaded))
+        bar.progress((fi + 1) / len(uploaded))
 
-    bar.empty(); status.empty()
-    st.session_state.excel = make_excel(st.session_state.records) \
-                             if st.session_state.records else None
-    st.session_state.done  = True
+    bar.empty()
+    status.empty()
 
+    if st.session_state.records:
+        st.session_state.excel = make_excel(st.session_state.records)
+
+    st.session_state.done = True
     elapsed = time.time() - t0
     st.success(
         f"Processed {len(st.session_state.records)} page(s) "
@@ -459,17 +372,22 @@ if run and uploaded:
 # RESULTS
 # ═══════════════════════════════════════════════════════════
 if st.session_state.done and st.session_state.records:
-    df = pd.DataFrame(st.session_state.records)
+    records = st.session_state.records
+    df      = pd.DataFrame(records)
+
+    # Display columns — exclude raw output from table
+    display_cols = [c for c in df.columns if c != "Raw Output"]
+    df_display   = df[display_cols]
 
-    # Stats strip
+    # Stats
     n_plots   = df["Plot Number"].astype(bool).sum()
-    n_amounts = df["Amount (KES)"].notna().sum()
+    n_amounts = pd.to_numeric(df["Amount (KES)"], errors="coerce").notna().sum()
     n_dates   = df["Date"].astype(bool).sum()
 
     st.markdown(f"""
     <div class="stats-strip">
       <div class="stat-item">
-        <div class="stat-n">{len(df)}</div>
+        <div class="stat-n">{len(records)}</div>
         <div class="stat-l">Pages processed</div>
       </div>
       <div class="stat-item">
@@ -487,7 +405,7 @@ if st.session_state.done and st.session_state.records:
     </div>
     """, unsafe_allow_html=True)
 
-    # Table header + download
+    # Table header + download side by side
     col_t, col_d = st.columns([5, 1])
     with col_t:
         st.markdown("""
@@ -505,9 +423,23 @@ if st.session_state.done and st.session_state.records:
                 mime="application/vnd.openxmlformats-officedocument.spreadsheetml.sheet",
             )
 
-    # Table
-    st.dataframe(df, use_container_width=True,
-                 height=min(80 + len(df)*38, 560), hide_index=True)
+    # Dataframe
+    st.dataframe(
+        df_display,
+        use_container_width=True,
+        height=min(80 + len(df) * 38, 560),
+        hide_index=True,
+    )
+
+    # Raw OCR output — for verification/debugging
+    with st.expander("🔍 View raw model output (for verification)"):
+        for r in records:
+            st.markdown(
+                f'<div style="font-family:monospace;font-size:0.78rem;'
+                f'padding:0.5rem 0;border-bottom:1px solid #E5E7EB;color:#374151">'
+                f'<strong>{r["File"]}</strong><br>{r.get("Raw Output","")}</div>',
+                unsafe_allow_html=True,
+            )
 
     # Errors
     if st.session_state.errors: