Spaces:
Running
Running
File size: 16,419 Bytes
b10aabd 353f247 b10aabd 353f247 b10aabd ce64eb7 b10aabd ce64eb7 b10aabd ce64eb7 b10aabd ce64eb7 b10aabd ce64eb7 b10aabd ce64eb7 b10aabd ce64eb7 b10aabd ce64eb7 b10aabd ce64eb7 b10aabd | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 | """Heuristic parser for LinkedIn PDF exports. Zero LLM calls, instant results.
LinkedIn PDFs use a two-column layout where the right column (skills, languages)
is interleaved line-by-line with the left column (experience). We handle this
by dumping everything between Experience and Education into one stream, then
filtering out right-column noise using known patterns.
"""
import re
from app.models import Education, Experience, Profile
# --- Patterns ---
DATE_LINE = re.compile(
r"^((?:jan|fév|fé|mar|avr|mai|jun|jui|aoû|ao|sep|oct|nov|déc|"
r"january|february|march|april|may|june|july|august|september|october|november|december|"
r"janvier|février|mars|avril|juin|juillet|août|septembre|octobre|novembre|décembre)"
r"\w*\.?\s+\d{4})"
r"\s*[-–—]\s*"
r"((?:jan|fév|fé|mar|avr|mai|jun|jui|aoû|ao|sep|oct|nov|déc|"
r"january|february|march|april|may|june|july|august|september|october|november|december|"
r"janvier|février|mars|avril|juin|juillet|août|septembre|octobre|novembre|décembre)"
r"\w*\.?\s+\d{4}|present|présent|aujourd'hui|current|now)"
r"\s*(\(.*\))?",
re.IGNORECASE,
)
# Right-column noise patterns
LANGUAGE_LINE = re.compile(
r"^(French|English|Spanish|German|Italian|Portuguese|Chinese|Japanese|Korean|Arabic|Russian|Dutch|Hindi|Mandarin|Cantonese|"
r"Français|Anglais|Espagnol|Allemand|Italien|Portugais|Chinois|Japonais|Coréen|Arabe|Russe|Néerlandais)"
r"\s*\(", re.IGNORECASE
)
PAGE_MARKER = re.compile(r"^Page \d+ of \d+$")
PHONE_RE = re.compile(r"(\+?\d[\d\s\-\.]{7,15}\d)")
EMAIL_RE = re.compile(r"[\w.+-]+@[\w-]+\.[\w.-]+")
LINKEDIN_RE = re.compile(r"((?:https?://)?(?:www\.)?linkedin\.com/in/[\w-]+)")
LOCATION_LINE = re.compile(
r"^(Paris|Lyon|Marseille|Toulouse|Nantes|Bordeaux|Lille|Strasbourg|London|Berlin|"
r"New York|San Francisco|Remote|France|Région de Paris|Paris Area|Île-de-France)",
re.IGNORECASE
)
SECTION_HEADERS = {
"experience": re.compile(r"^(Experience|Expérience)$", re.IGNORECASE),
"education": re.compile(r"^(Education|Formation)$", re.IGNORECASE),
"skills": re.compile(r"^(Skills|Compétences|Top Skills|Principales compétences)$", re.IGNORECASE),
"languages": re.compile(r"^(Languages|Langues)$", re.IGNORECASE),
"summary": re.compile(r"^(Summary|Résumé|About|À propos)$", re.IGNORECASE),
"certifications": re.compile(r"^(Certifications|Licences|Licenses)$", re.IGNORECASE),
"honors": re.compile(r"^(Honors|Distinctions|Honors & Awards)$", re.IGNORECASE),
"volunteer": re.compile(r"^(Volunteer Experience|Bénévolat)$", re.IGNORECASE),
}
# Sections that end the experience block
END_EXPERIENCE_SECTIONS = {"education", "certifications", "honors", "volunteer"}
def _section_type(line: str) -> str | None:
for name, pat in SECTION_HEADERS.items():
if pat.match(line.strip()):
return name
return None
def _is_right_column_noise(line: str) -> bool:
"""Check if a line is right-column noise (language proficiency, etc.)."""
stripped = line.strip()
if LANGUAGE_LINE.match(stripped):
return True
if stripped in ("(LinkedIn)", "(Mobile)"):
return True
if PAGE_MARKER.match(stripped):
return True
# Duration-only line like "4 ans 3 mois" (multi-role company header)
if re.match(r"^\d+\s+(an|mois|year|month)", stripped, re.IGNORECASE):
return False # Not noise — it's a company group marker
return False
def _is_location_line(line: str) -> bool:
return bool(LOCATION_LINE.match(line.strip()))
def parse_linkedin_heuristic(raw_text: str) -> Profile | None:
"""Parse LinkedIn PDF text into a Profile using heuristics.
Returns None if it doesn't look like a LinkedIn PDF.
"""
lines = [l.strip() for l in raw_text.split("\n") if l.strip()]
if len(lines) < 5:
return None
# --- Phase 1: Find section boundaries ---
exp_start = None
exp_end = None
edu_start = None
skills_start = None
summary_start = None
for i, line in enumerate(lines):
sec = _section_type(line)
if sec == "experience" and exp_start is None:
exp_start = i + 1
elif sec in END_EXPERIENCE_SECTIONS and exp_start is not None and exp_end is None:
exp_end = i
if sec == "education":
edu_start = i + 1
elif sec == "skills" and exp_start is None:
skills_start = i + 1
elif sec == "summary" and exp_start is None:
summary_start = i + 1
if exp_start is None:
return None
if exp_end is None:
exp_end = len(lines)
# --- Phase 2: Parse header (everything before experience) ---
header_lines = lines[:exp_start - 1]
name, email, phone, linkedin, location, summary = _parse_header(
header_lines, lines[summary_start:exp_start - 1] if summary_start else []
)
# --- Phase 3: Parse skills (before experience) ---
skills: list[str] = []
if skills_start:
for line in lines[skills_start:exp_start - 1]:
sec = _section_type(line)
if sec:
break
stripped = line.strip()
if stripped and not _is_right_column_noise(stripped):
skills.append(stripped)
# --- Phase 4: Parse experience block ---
# Filter right-column noise, then extract experiences
exp_raw = lines[exp_start:exp_end]
# Collect languages from interleaved right-column
languages: list[str] = []
exp_clean: list[str] = []
for line in exp_raw:
stripped = line.strip()
sec = _section_type(stripped)
if sec in ("skills", "languages"):
continue # Skip right-column section headers
if PAGE_MARKER.match(stripped):
continue
if _is_right_column_noise(stripped):
languages.append(stripped)
continue
exp_clean.append(stripped)
experiences = _parse_experiences(exp_clean)
# --- Phase 5: Parse education ---
education: list[Education] = []
if edu_start:
edu_lines = lines[edu_start:]
# Stop at next section or end
edu_clean = []
for line in edu_lines:
sec = _section_type(line)
if sec and sec != "education":
break
if not PAGE_MARKER.match(line.strip()):
edu_clean.append(line.strip())
education = _parse_education(edu_clean)
# Title = first experience title if available
title = experiences[0].title if experiences else ""
profile = Profile(
name=name,
title=title,
location=location,
email=email,
phone=phone,
linkedin=linkedin,
summary=summary,
experiences=experiences,
education=education,
skills=skills,
languages=languages,
)
if profile.name and len(profile.experiences) > 0 and _is_coherent(profile, raw_text):
return profile
return None
def _is_coherent(profile: Profile, raw_text: str) -> bool:
"""Quick sanity check — if the parse looks off, return False to trigger LLM fallback."""
# Name should be 2+ words (not a section header or label)
if len(profile.name.split()) < 2:
return False
# At least 30% of experiences should have a company name
if profile.experiences:
with_company = sum(1 for e in profile.experiences if e.company)
if with_company / len(profile.experiences) < 0.3:
return False
# Experiences count should be plausible relative to PDF size
# A typical LinkedIn PDF has ~500 chars per experience
expected_min = max(1, len(raw_text) // 2000)
if len(profile.experiences) < expected_min // 2:
return False
# Name should appear somewhere in the raw text
if profile.name.lower() not in raw_text.lower():
return False
return True
def _parse_header(header_lines: list[str], summary_lines: list[str]) -> tuple:
"""Extract contact info from header."""
name = ""
email = ""
phone = ""
linkedin = ""
location = ""
summary = ""
all_text = "\n".join(header_lines)
m = EMAIL_RE.search(all_text)
if m:
email = m.group()
m = PHONE_RE.search(all_text)
if m:
phone = m.group().strip()
m = LINKEDIN_RE.search(all_text)
if m:
linkedin = m.group()
for line in header_lines:
stripped = line.strip()
if stripped in ("(LinkedIn)", "(Mobile)") or PAGE_MARKER.match(stripped):
continue
if EMAIL_RE.search(stripped) or LINKEDIN_RE.search(stripped):
continue
# Clean "Coordonnées" prefix
cleaned = stripped
for prefix in ("Coordonnées ", "Contact "):
if cleaned.startswith(prefix):
cleaned = cleaned[len(prefix):]
# Phone-only line
if PHONE_RE.match(cleaned) and cleaned.replace(" ", "").replace("+", "").replace("-", "").replace(".", "").isdigit():
continue
if not name and cleaned and len(cleaned) < 60:
name = cleaned
elif not location and _is_location_line(cleaned):
location = cleaned
if summary_lines:
summary = " ".join(l.strip() for l in summary_lines if l.strip() and not _section_type(l))
return name, email, phone, linkedin, location, summary
DUR_PATTERN = re.compile(r"^\d+\s+(an|mois|year|month)", re.IGNORECASE)
def _find_title_company_before_date(lines: list[str], date_idx: int) -> tuple[str, str, bool]:
"""Look backwards from a date line to find the title, company, and whether this is a multi-role group.
Returns (company, title, is_multi_role).
"""
# Collect non-location, non-bullet lines going backwards from the date
candidates = []
for i in range(date_idx - 1, max(date_idx - 5, -1), -1):
line = lines[i].strip()
if not line:
continue
if DATE_LINE.match(line):
break # Hit previous date — stop
if _is_location_line(line):
continue
if line.startswith(("- ", "* ", "• ", "· ")):
continue # Skip bullets — they belong to the previous experience
candidates.insert(0, line)
# Check for duration line (multi-role indicator)
dur_idx = None
for ci, c in enumerate(candidates):
if DUR_PATTERN.match(c):
dur_idx = ci
break
if dur_idx is not None and dur_idx > 0:
# Multi-role: Company / Duration / Title / Date
company = candidates[dur_idx - 1]
title_candidates = candidates[dur_idx + 1:]
title = title_candidates[0] if title_candidates else ""
return company, title, True
if len(candidates) >= 2:
return candidates[-2], candidates[-1], False
if len(candidates) == 1:
return candidates[0], "", False
return "", "", False
def _parse_experiences(lines: list[str]) -> list[Experience]:
"""Two-pass parser: find dates first, then look backwards for metadata and forwards for bullets."""
# Pass 1: Find all date lines and extract company/title by looking backwards
entries: list[dict] = []
for i, line in enumerate(lines):
if DATE_LINE.match(line.strip()):
company, title, is_multi = _find_title_company_before_date(lines, i)
entries.append({
"idx": i,
"company": company,
"title": title,
"is_multi": is_multi,
"dates": re.sub(r"\s*\(.*?\)\s*$", "", line.strip()),
})
# Propagate group_company for multi-role entries
# When LinkedIn shows multiple roles at one company, only the first has
# Company + Duration. Subsequent roles only show Title + Date.
# The lookback will mistake the title for a company (single candidate).
group_company = None
for e in entries:
if e["is_multi"]:
group_company = e["company"]
elif group_company:
# The detected "company" is likely a job title (only 1 candidate in lookback)
# Swap: what we thought was company is actually the title
if e["company"] and not e["title"]:
e["title"] = e["company"]
e["company"] = group_company
else:
group_company = None
# Pass 2: Collect bullets between consecutive dates
# Build a set of "metadata lines" (company/title) to exclude from bullets
meta_lines = set()
for e in entries:
if e["company"]:
meta_lines.add(e["company"])
if e["title"]:
meta_lines.add(e["title"])
experiences: list[Experience] = []
for di, e in enumerate(entries):
start = e["idx"] + 1
end = entries[di + 1]["idx"] if di + 1 < len(entries) else len(lines)
bullets = []
for i in range(start, end):
line = lines[i].strip()
if not line or _is_location_line(line) or DUR_PATTERN.match(line):
continue
if line in meta_lines:
continue
if line.startswith(("- ", "* ", "• ", "· ")):
bullets.append(line.lstrip("-*•· ").strip())
elif len(line) > 10:
bullets.append(line)
experiences.append(Experience(
title=e["title"],
company=e["company"],
dates=e["dates"],
description=" ".join(bullets[:3]),
bullets=bullets,
))
return _merge_same_company(experiences)
def _merge_same_company(experiences: list[Experience]) -> list[Experience]:
"""Merge consecutive experiences at the same company into a single entry with combined bullets."""
if not experiences:
return experiences
merged: list[Experience] = []
for exp in experiences:
if (merged
and exp.company
and merged[-1].company
and exp.company.lower() == merged[-1].company.lower()):
# Same company — merge into previous
prev = merged[-1]
# Combine titles
if exp.title and exp.title.lower() != prev.title.lower():
combined_title = f"{prev.title} → {exp.title}"
else:
combined_title = prev.title
# Combine dates (earliest start - latest end)
combined_dates = f"{exp.dates} / {prev.dates}" if exp.dates != prev.dates else prev.dates
# Combine bullets, prefixed with role title
combined_bullets = []
if prev.bullets:
combined_bullets.append(f"[{prev.title}]")
combined_bullets.extend(prev.bullets)
if exp.bullets:
combined_bullets.append(f"[{exp.title}]")
combined_bullets.extend(exp.bullets)
merged[-1] = Experience(
title=combined_title,
company=prev.company,
dates=combined_dates,
description=prev.description,
bullets=combined_bullets if combined_bullets else prev.bullets + exp.bullets,
)
else:
merged.append(exp)
return merged
def _build_exp(data: dict) -> Experience:
bullets = data.get("bullets", [])
desc = " ".join(bullets[:3]) if bullets else ""
return Experience(
title=data.get("title", ""),
company=data.get("company", ""),
dates=data.get("dates", ""),
description=desc,
bullets=bullets,
)
def _parse_education(lines: list[str]) -> list[Education]:
"""Parse education lines. Format: School / Degree, Field · (2011 - 2015)"""
entries: list[Education] = []
i = 0
while i < len(lines):
school = lines[i].strip()
degree = ""
year = ""
if i + 1 < len(lines):
next_line = lines[i + 1].strip()
year_match = re.search(r"\((\d{4}\s*[-–]\s*\d{4}|\d{4})\)", next_line)
if year_match:
year = year_match.group(1)
degree = next_line[:year_match.start()].rstrip(" ·,")
i += 2
else:
degree = next_line
i += 2
else:
i += 1
if school:
entries.append(Education(school=school, degree=degree, year=year))
return entries
|