cd /tmp/workspace/attachments/ad798716-f03a-4471-a851-67a398745def && python3 -c " import pdfplumber, os files = { 'paper1': 'paper 1.pdf', 'paper2': 'paper 2.pdf', 'paper3': 'Paper 3.pdf', 'paper4': 'Paper 4.pdf' } for key, fname in files.items(): print(f'\n\n====== {key}: {fname} ======') with pdfplumber.open(fname) as pdf: for i, page in enumerate(pdf.pages): text = page.extract_text() print(f'--- Page {i+1} ---') print(text) "

cd /tmp/workspace/attachments/ad798716-f03a-4471-a851-67a398745def && python3 -c " from pypdf import PdfReader import os for fname in ['paper 1.pdf', 'paper 2.pdf', 'Paper 3.pdf', 'Paper 4.pdf']: reader = PdfReader(fname) print(f'{fname}: {len(reader.pages)} pages') # Try to get text from each page for i, page in enumerate(reader.pages): t = page.extract_text() if t and t.strip(): print(f' Page {i+1}: {t[:200]}') else: print(f' Page {i+1}: [no text layer]') "

python3 -c " from pypdf import PdfReader import subprocess, os # Check if pdf2image or similar is available try: from pdf2image import convert_from_path print('pdf2image available') except: print('pdf2image NOT available') try: import fitz # PyMuPDF print('PyMuPDF (fitz) available') except: print('PyMuPDF NOT available') try: import PIL print('PIL available:', PIL.__version__) except: print('PIL NOT available') "

mkdir -p /tmp/workspace/paper_images && cd /tmp/workspace/attachments/ad798716-f03a-4471-a851-67a398745def && python3 -c " from pdf2image import convert_from_path import os base_out = '/tmp/workspace/paper_images' files = { 'paper1': 'paper 1.pdf', 'paper2': 'paper 2.pdf', 'paper3': 'Paper 3.pdf', 'paper4': 'Paper 4.pdf' } for key, fname in files.items(): pages = convert_from_path(fname, dpi=150) for i, page in enumerate(pages): out_path = f'{base_out}/{key}_page{i+1}.png' page.save(out_path, 'PNG') print(f'Saved {out_path}') print('Done') " 2>&1

cd /tmp/workspace/paper_images && python3 -m http.server 8765 &>/tmp/http_server.log & echo "Server PID: $!" && sleep 2 && curl -s -o /dev/null -w "%{http_code}" http://localhost:8765/paper1_page1.png

python3 -c " from PIL import Image import os base = '/tmp/workspace/paper_images' # Combine all 8 pages of each paper into a single tall image for paper in ['paper1', 'paper2', 'paper3', 'paper4']: images = [] for i in range(1, 9): path = f'{base}/{paper}_page{i}.png' img = Image.open(path) images.append(img) # Resize each page to width 800 maintaining aspect ratio resized = [] for img in images: w, h = img.size new_w = 800 new_h = int(h * new_w / w) resized.append(img.resize((new_w, new_h), Image.LANCZOS)) total_height = sum(img.height for img in resized) combined = Image.new('RGB', (800, total_height), 'white') y = 0 for img in resized: combined.paste(img, (0, y)) y += img.height out_path = f'{base}/{paper}_combined.png' combined.save(out_path, 'PNG', optimize=True) size = os.path.getsize(out_path) print(f'{paper}_combined.png: {size/1024/1024:.2f} MB, {total_height}px tall') "

python3 -c " # Try markitdown try: from markitdown import MarkItDown md = MarkItDown() result = md.convert('/tmp/workspace/attachments/ad798716-f03a-4471-a851-67a398745def/paper 1.pdf') print('markitdown result:') print(result.text_content[:3000]) except Exception as e: print('markitdown error:', e) "

# Try tesseract OCR tesseract --version 2>&1 || echo "tesseract not available" # Try easyocr python3 -c "import easyocr; print('easyocr available')" 2>&1 || echo "easyocr not available" # Try paddleocr python3 -c "import paddleocr; print('paddleocr available')" 2>&1 || echo "paddleocr not available" # Check what vision libs available python3 -c "import google.cloud.vision; print('google vision')" 2>&1 || echo "no google vision" python3 -c "import anthropic; print('anthropic')" 2>&1 || echo "no anthropic" python3 -c "import openai; print('openai available')" 2>&1

python3 -c " import openai, base64, os # Read API key from environment api_key = os.environ.get('OPENAI_API_KEY', '') print('API key available:', bool(api_key)) # Try to use OpenAI vision API to read the combined images client = openai.OpenAI() # Read paper1 combined image as base64 with open('/tmp/workspace/paper_images/paper1_combined.png', 'rb') as f: img_data = base64.b64encode(f.read()).decode('utf-8') response = client.chat.completions.create( model='gpt-4o', messages=[ { 'role': 'user', 'content': [ { 'type': 'image_url', 'image_url': { 'url': f'data:image/png;base64,{img_data}', 'detail': 'high' } }, { 'type': 'text', 'text': 'This is a surgical exam question paper. Extract ALL questions exactly as written, with question numbers, sub-parts (a, b, c), and marks allocation. Be complete and precise.' } ] } ], max_tokens=4000 ) print(response.choices[0].message.content) " 2>&1 | head -200

# Try installing tesseract via pip (pytesseract uses system tesseract, but maybe there's a pure python one) pip install --user pytesseract 2>&1 | tail -3 # Try to install tesseract via pip wrapper pip install --user tesserocr 2>&1 | tail -5 # try ocrmypdf pip install --user ocrmypdf 2>&1 | tail -3

# Try tesserocr directly python3 -c " import tesserocr print('tesserocr version:', tesserocr.__version__) from PIL import Image img = Image.open('/tmp/workspace/paper_images/paper1_page1.png') text = tesserocr.image_to_text(img) print(text[:2000]) " 2>&1

https://drive.google.com/drive/folders/1LGxtuq73u3qxC1tdwxUm…

mkdir -p /tmp/workspace/questions && cd /tmp/workspace/questions # Folder 1 (link 1): MSGS101-104 (Oct 2025) # Folder 2 (link 2): MSGS101-104 (Jan 2025) # Folder 3 (link 3): GENERAL SURGERY subfolders # First try to get file IDs from folder 1 - MSGS101.doc # Try direct download using folder ID approach # Folder 1 ID: 1LGxtuq73u3qxC1tdwxUm-q9caGGUq6xW # Folder 2 ID: 1lCa5KNDM1-yD0JuuK15M7Aq71kLpEWOg # Folder 3 ID: 1UUkrC5HWzkvAfxCNktuoPduDKpwOF6mz # Try to download folder contents via gdown pip install --user gdown 2>&1 | tail -3

cd /tmp/workspace/questions export PATH="$PATH:/opt/amazon/genesis1p-tools/var/.local/bin" # Download folder 1 (Oct 2025 - General Surgery) gdown --folder "https://drive.google.com/drive/folders/1LGxtuq73u3qxC1tdwxUm-q9caGGUq6xW" -O /tmp/workspace/questions/folder1 2>&1 | tail -20

export PATH="$PATH:/opt/amazon/genesis1p-tools/var/.local/bin" # Download folder 2 (Jan 2025) gdown --folder "https://drive.google.com/drive/folders/1lCa5KNDM1-yD0JuuK15M7Aq71kLpEWOg" -O /tmp/workspace/questions/folder2 2>&1 | tail -10 echo "---" # Download folder 3 (GENERAL SURGERY with subfolders) gdown --folder "https://drive.google.com/drive/folders/1UUkrC5HWzkvAfxCNktuoPduDKpwOF6mz" -O /tmp/workspace/questions/folder3 2>&1 | tail -20

ls -la /tmp/workspace/questions/folder1/ ls -la /tmp/workspace/questions/folder2/ ls -la /tmp/workspace/questions/folder3/ 2>/dev/null find /tmp/workspace/questions/folder3 -name "*.doc" 2>/dev/null | head -20

find /tmp/workspace/questions/folder3 -name "*.doc" | sort echo "---" find /tmp/workspace/questions/folder3 -name "*.doc" | wc -l

python3 -c " import subprocess, os # .doc files are legacy binary - try reading with python-docx (it sometimes works) from docx import Document test_file = '/tmp/workspace/questions/folder3/Basic Sciences as Applied to Surgery/MSGS101 (2017).doc' try: doc = Document(test_file) print('python-docx worked!') for p in doc.paragraphs: if p.text.strip(): print(repr(p.text)) except Exception as e: print('python-docx failed:', e) # Try raw binary extraction - .doc files often have text accessible via strings result = subprocess.run(['strings', '-n', '4', test_file], capture_output=True, text=True, errors='replace') if result.returncode == 0: lines = [l for l in result.stdout.split('\n') if len(l) > 10 and not l.startswith('\x0c')] print('--- strings output (first 100 lines) ---') for line in lines[:100]: print(line) " 2>&1 | head -150

python3 << 'EOF' import subprocess, os, re # Try several approaches on the .doc file test_file = '/tmp/workspace/questions/folder3/Basic Sciences as Applied to Surgery/MSGS101 (2017).doc' # Method 1: antiword try: result = subprocess.run(['antiword', test_file], capture_output=True, text=True, errors='replace') print('antiword:', result.stdout[:500] if result.stdout else 'no output', result.stderr[:200]) except: print('antiword not found') # Method 2: catdoc try: result = subprocess.run(['catdoc', test_file], capture_output=True, text=True, errors='replace') print('catdoc:', result.stdout[:500] if result.stdout else 'no output', result.stderr[:200]) except: print('catdoc not found') # Method 3: wvText try: result = subprocess.run(['wvText', test_file, '/dev/stdout'], capture_output=True, text=True, errors='replace') print('wvText:', result.stdout[:500] if result.stdout else 'no output') except: print('wvText not found') # Method 4: use python-oletools try: import olefile ole = olefile.OleFileIO(test_file) print('olefile streams:', ole.listdir()) except Exception as e: print('olefile error:', e) EOF

pip install --user olefile 2>&1 | tail -2 python3 -c " import olefile test_file = '/tmp/workspace/questions/folder3/Basic Sciences as Applied to Surgery/MSGS101 (2017).doc' ole = olefile.OleFileIO(test_file) print('Streams:', ole.listdir()) # WordDocument stream if ole.exists('WordDocument'): data = ole.openstream('WordDocument').read() print('WordDocument stream size:', len(data)) if ole.exists('1Table') or ole.exists('0Table'): tbl = ole.openstream('1Table' if ole.exists('1Table') else '0Table').read() print('Table stream size:', len(tbl)) "

ls "/tmp/workspace/questions/folder3/Basic Sciences as Applied to Surgery/" | head -5 # Try with escaped path python3 << 'EOF' import olefile, glob, os base = '/tmp/workspace/questions/folder3' files = glob.glob(base + '/**/*.doc', recursive=True) print(f'Found {len(files)} .doc files') print('First file:', repr(files[0])) # Try olefile on first file try: ole = olefile.OleFileIO(files[0]) print('Streams:', [str(s) for s in ole.listdir()]) except Exception as e: print('olefile error:', type(e).__name__, e) EOF

ls /tmp/workspace/questions/folder3/ echo "---" ls /tmp/workspace/questions/folder3/* | head -30 echo "---" find /tmp/workspace/questions/folder3 -name "*.doc" 2>/dev/null | head -5

ls /tmp/workspace/questions/ ls /tmp/workspace/

mkdir -p /tmp/workspace/questions && export PATH="$PATH:/opt/amazon/genesis1p-tools/var/.local/bin" # Download folder 3 which has all years organized by subject gdown --folder "https://drive.google.com/drive/folders/1UUkrC5HWzkvAfxCNktuoPduDKpwOF6mz" -O /tmp/workspace/questions/folder3 2>&1 | grep -E "(Download|To:|Failed|Error)" | head -50

export PATH="$PATH:/opt/amazon/genesis1p-tools/var/.local/bin" gdown --folder "https://drive.google.com/drive/folders/1UUkrC5HWzkvAfxCNktuoPduDKpwOF6mz" -O /tmp/workspace/questions/folder3 2>&1 | tail -30 ls /tmp/workspace/questions/

pip install --user gdown 2>&1 | tail -2 && python3 -m gdown --version

mkdir -p /tmp/workspace/questions/folder3 python3 -m gdown --folder "https://drive.google.com/drive/folders/1UUkrC5HWzkvAfxCNktuoPduDKpwOF6mz" -O /tmp/workspace/questions/folder3 2>&1 | tail -40

find /tmp/workspace/questions -name "*.doc" 2>/dev/null | sort | head -40 echo "---" find /tmp/workspace/questions -name "*.doc" 2>/dev/null | wc -l

pip install --user olefile 2>&1 | tail -2 python3 << 'PYEOF' import olefile, struct, glob, os, re def extract_doc_text(filepath): """Extract text from legacy .doc file using olefile.""" try: ole = olefile.OleFileIO(filepath) # Read WordDocument stream if not ole.exists('WordDocument'): return "[No WordDocument stream]" wd = ole.openstream('WordDocument').read() # FIB (File Information Block) - get character positions # Magic bytes check magic = struct.unpack_from('<H', wd, 0)[0] # Table stream name table_name = '1Table' if struct.unpack_from('<H', wd, 0x000A)[0] & 0x0200 else '0Table' if not ole.exists(table_name): table_name = '1Table' if ole.exists('1Table') else '0Table' if not ole.exists(table_name): # Fall back to raw text extraction text = wd.decode('latin-1', errors='replace') # Filter printable characters lines = [] current = [] for ch in text: if ch.isprintable() or ch in '\n\r\t': current.append(ch) elif current: word = ''.join(current).strip() if len(word) > 3: lines.append(word) current = [] return '\n'.join(lines) table = ole.openstream(table_name).read() # Get text using clx structure # FibRgFcLcb97 - fcClx offset at 0x01A2, lcbClx at 0x01A6 fc_clx = struct.unpack_from('<I', wd, 0x01A2)[0] lcb_clx = struct.unpack_from('<I', wd, 0x01A6)[0] if fc_clx >= len(table) or lcb_clx == 0: # Simple extraction from main stream fc_min = struct.unpack_from('<I', wd, 0x0018)[0] # ccpText # Just try to decode raw wd stream raw = wd[fc_min:].decode('latin-1', errors='replace') return raw[:5000] # Get text from WordDocument stream directly # ccpText field at offset 0x004C ccp_text = struct.unpack_from('<I', wd, 0x004C)[0] # fcMin - start of text in WordDocument fc_min = struct.unpack_from('<I', wd, 0x0018)[0] # Try simple extraction text_bytes = wd[fc_min:fc_min + ccp_text * 2] # Could be Unicode (2 bytes per char) try: text = text_bytes.decode('utf-16-le', errors='replace') except: text = text_bytes.decode('latin-1', errors='replace') # Clean up text = re.sub(r'[\x00-\x08\x0b\x0c\x0e-\x1f\x7f-\x9f]', ' ', text) text = re.sub(r' +', ' ', text) return text.strip() except Exception as e: return f"[Error: {e}]" # Test on first file test_file = sorted(glob.glob('/tmp/workspace/questions/folder3/**/*.doc', recursive=True))[0] print(f"Testing: {test_file}") print(extract_doc_text(test_file)[:3000]) PYEOF

pip install --user docx2txt 2>&1 | tail -2 # docx2txt works on .docx not .doc # Try msoffcrypto and doc2docx type approach # Check if file is actually a .docx disguised as .doc python3 << 'PYEOF' import zipfile, olefile, glob files = sorted(glob.glob('/tmp/workspace/questions/folder3/**/*.doc', recursive=True)) # Check if any are actually zip/docx for f in files[:4]: try: z = zipfile.ZipFile(f) print(f"{f.split('/')[-1]}: IS A ZIP/DOCX") z.close() except: pass try: ole = olefile.OleFileIO(f) streams = ole.listdir() # Get all stream names stream_names = ['/'.join(s) for s in streams] print(f"{f.split('/')[-1]}: OLE streams: {stream_names[:5]}") except Exception as e: print(f"{f.split('/')[-1]}: Error - {e}") PYEOF

python3 << 'PYEOF' from docx import Document import glob, os files = sorted(glob.glob('/tmp/workspace/questions/folder3/**/*.doc', recursive=True)) print(f"Total files: {len(files)}") # Test reading first file f = files[0] print(f"\n=== {f.split('/')[-1]} ===") try: doc = Document(f) for p in doc.paragraphs: if p.text.strip(): print(p.text) except Exception as e: print(f"Error: {e}") PYEOF

python3 << 'PYEOF' import zipfile, glob # These are ZIP files - let's look at content f = '/tmp/workspace/questions/folder3/Basic Sciences as Applied to Surgery/MSGS101 (2017).doc' with zipfile.ZipFile(f) as z: names = z.namelist() print("ZIP contents:", names[:20]) # Look for document.xml for name in names: if 'document' in name.lower() or '.xml' in name.lower(): print(f"\n--- {name} ---") content = z.read(name).decode('utf-8', errors='replace') print(content[:2000]) break PYEOF

python3 << 'PYEOF' import zipfile, glob, os # Check all files for folder in ['folder1', 'folder2']: path = f'/tmp/workspace/questions/{folder}' if not os.path.exists(path): print(f"{folder}: NOT EXISTS - need to re-download") continue files = glob.glob(f'{path}/*.doc') print(f"\n{folder}: {len(files)} files") if files: f = files[0] print(f" Testing: {f}") with zipfile.ZipFile(f) as z: names = z.namelist() print(f" Contents: {names[:10]}") # find document.xml doc_xml = [n for n in names if 'document.xml' in n or 'word/document' in n.lower()] print(f" Doc XML: {doc_xml}") if doc_xml: content = z.read(doc_xml[0]).decode('utf-8', errors='replace') # Strip XML tags to get text import re text = re.sub(r'<[^>]+>', ' ', content) text = re.sub(r'\s+', ' ', text).strip() print(f" Text preview: {text[:500]}") PYEOF # Also re-download folder1 and folder2 python3 -m gdown --folder "https://drive.google.com/drive/folders/1LGxtuq73u3qxC1tdwxUm-q9caGGUq6xW" -O /tmp/workspace/questions/folder1 2>&1 | grep -E "(To:|Download)" | head -10 & python3 -m gdown --folder "https://drive.google.com/drive/folders/1lCa5KNDM1-yD0JuuK15M7Aq71kLpEWOg" -O /tmp/workspace/questions/folder2 2>&1 | grep -E "(To:|Download)" | head -10 & wait echo "Downloads done"

ls /tmp/workspace/questions/ ls /tmp/workspace/questions/folder1/ 2>/dev/null || echo "folder1 missing" ls /tmp/workspace/questions/folder2/ 2>/dev/null || echo "folder2 missing" python3 << 'PYEOF' import zipfile, glob # Check folder3 files more carefully files = sorted(glob.glob('/tmp/workspace/questions/folder3/**/*.doc', recursive=True)) print(f"Folder3: {len(files)} files") # Check each one for f in files[:3]: print(f"\n{f.split('/')[-1]}:") try: with zipfile.ZipFile(f) as z: names = z.namelist() has_doc = any('word/document' in n.lower() for n in names) print(f" Has word/document.xml: {has_doc}") print(f" All files: {names}") except Exception as e: print(f" Error: {e}") PYEOF

python3 << 'PYEOF' import zipfile, glob for folder in ['folder1', 'folder2']: files = sorted(glob.glob(f'/tmp/workspace/questions/{folder}/*.doc')) print(f"\n{folder}: {len(files)} files") for f in files[:2]: print(f" {f.split('/')[-1]}:") try: with zipfile.ZipFile(f) as z: names = z.namelist() has_doc = any('word/document' in n.lower() for n in names) print(f" Has word/document.xml: {has_doc}, files: {names[:8]}") if has_doc: import re content = z.read('word/document.xml').decode('utf-8', errors='replace') text = re.sub(r'<[^>]+>', ' ', content) text = re.sub(r'\s+', ' ', text).strip() print(f" Text: {text[:400]}") except Exception as e: print(f" Error: {e}") PYEOF