File size: 19,294 Bytes
c643b04
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
from __future__ import annotations

import hashlib
import math
import re
from datetime import datetime

from .narrative import career_text, current_role, full_text, profile, summary_text

PROF_WEIGHT = {"beginner": 0.30, "intermediate": 0.60, "advanced": 0.85, "expert": 1.00}
TIER_ORDER = {"trap": 0, "weak": 1, "possible": 2, "strong": 3}
AI_SKILL_TERMS = (
    "machine learning", "deep learning", "nlp", "pytorch", "tensorflow", "llm", "rag",
    "vector", "embedding", "fine-tun", "lora", "qlora", "faiss", "ranking",
    "recommendation", "search", "retrieval", "python", "evaluation",
)
ASSESSMENT_TERMS = (
    "machine learning", "deep learning", "nlp", "pytorch", "tensorflow", "llm", "rag",
    "vector", "embedding", "ranking", "retrieval", "python", "evaluation", "search",
)
CV_SPEECH_ROBOTICS = ("computer vision", "image classification", "speech", "tts", "robotics", "autonomous")
DEFAULT_CONSULTING = (
    "tcs", "tata consultancy", "infosys", "wipro", "accenture", "cognizant", "capgemini",
    "hcl", "tech mahindra", "mindtree", "ltimindtree", "mphasis", "hexaware",
)
STARTUP_SIZES = {"1-10", "11-50", "51-200", "201-500"}


def _lower(x: object) -> str:
    return str(x or "").lower()


def _year(date_text: object) -> int | None:
    m = re.search(r"\d{4}", str(date_text or ""))
    return int(m.group(0)) if m else None


def _title_tier(title: str, spec: dict) -> tuple[str, float]:
    tl = _lower(title)
    tiers = spec.get("title_tiers") or {}
    for trap in tiers.get("trap", []):
        if _lower(trap) in tl:
            return "trap", 0.0
    for strong in tiers.get("strong", []):
        if _lower(strong) in tl:
            return "strong", 1.0
    for possible in tiers.get("possible", []):
        if _lower(possible) in tl:
            return "possible", 0.70
    if any(k in tl for k in ("machine learning", " ml", " ai ", "ai engineer", "nlp", "search", "recommend", "data scien")):
        return "possible", 0.65
    return "weak", 0.35


def _best_title_tier(candidate: dict, spec: dict) -> tuple[str, float, str]:
    best_tier, best_score, best_title = "weak", 0.35, str(profile(candidate).get("current_title") or "")
    for role in candidate.get("career_history") or []:
        title = str(role.get("title") or "")
        tier, base = _title_tier(title, spec)
        months = int(role.get("duration_months") or 0)
        weighted = base * min(1.0, 0.35 + months / 30.0)
        if TIER_ORDER[tier] > TIER_ORDER[best_tier] or (
            TIER_ORDER[tier] == TIER_ORDER[best_tier] and weighted > best_score
        ):
            best_tier, best_score, best_title = tier, weighted, title
    cur_title = str(profile(candidate).get("current_title") or "")
    cur_tier, cur_score = _title_tier(cur_title, spec)
    if TIER_ORDER[cur_tier] >= TIER_ORDER[best_tier]:
        return cur_tier, max(best_score, cur_score), cur_title or best_title
    return best_tier, best_score, best_title


def _family_score(text: str, terms: list[str], hits_out: list[str], cap_hits: int = 3) -> float:
    hits = [term for term in terms if _lower(term) in text]
    hits_out.extend(hits[:3])
    return min(1.0, len(hits) / cap_hits)


def career_evidence(text: str, spec: dict) -> tuple[float, list[str]]:
    tl = _lower(text)
    families = spec.get("career_evidence_terms") or {}
    weights = {"retrieval": 0.28, "ranking": 0.28, "production": 0.24, "llm": 0.12, "plain_language": 0.08}
    hits: list[str] = []
    score = 0.0
    for family, weight in weights.items():
        score += weight * _family_score(tl, [str(x) for x in families.get(family, [])], hits)
    return min(1.0, score), list(dict.fromkeys(hits))[:6]


def _blended_evidence(candidate: dict, spec: dict) -> tuple[float, list[str], float, float]:
    ctext = career_text(candidate)
    stext = summary_text(candidate)
    career_score, career_hits = career_evidence(ctext, spec)
    summary_score, summary_hits = career_evidence(stext, spec)
    combined = min(1.0, 0.82 * career_score + 0.18 * summary_score)
    hits = list(dict.fromkeys([*career_hits, *summary_hits]))[:6]
    return combined, hits, career_score, summary_score


def skill_trust(candidate: dict) -> tuple[float, int, list[str]]:
    values: list[float] = []
    ai_values: list[float] = []
    ai_names: list[str] = []
    for skill in candidate.get("skills") or []:
        name = str(skill.get("name") or "")
        nl = name.lower()
        prof = PROF_WEIGHT.get(_lower(skill.get("proficiency")), 0.40)
        endorsements = int(skill.get("endorsements") or 0)
        months = float(skill.get("duration_months") or 0)
        trust = prof * math.log1p(endorsements) * min(months, 48) / 48.0
        trust = min(1.0, trust / 3.0)
        values.append(trust)
        if any(term in nl for term in AI_SKILL_TERMS):
            ai_values.append(trust)
            ai_names.append(name)
    if ai_values:
        return sum(ai_values) / len(ai_values), len(ai_values), ai_names[:6]
    return (sum(values) / len(values) if values else 0.0), 0, []


def assessment_score(candidate: dict) -> float:
    scores = (candidate.get("redrob_signals") or {}).get("skill_assessment_scores") or {}
    vals = [float(v) for k, v in scores.items() if any(term in _lower(k) for term in ASSESSMENT_TERMS)]
    if not vals:
        return 0.0
    return max(0.0, min(1.0, sum(vals) / len(vals) / 100.0))


def _education_score(candidate: dict) -> float:
    edu = candidate.get("education") or []
    if not edu:
        return 0.25
    score = 0.0
    for item in edu:
        field = _lower(item.get("field_of_study"))
        degree = _lower(item.get("degree"))
        tier = _lower(item.get("tier"))
        if any(x in field for x in ("computer", "machine learning", "artificial", "data science", "software", "electrical")):
            score += 0.35
        if any(x in degree for x in ("m.tech", "m.s", "ms", "phd", "master")):
            score += 0.20
        score += {"tier_1": 0.25, "tier_2": 0.18, "tier_3": 0.10, "tier_4": 0.05}.get(tier, 0.05)
    return min(1.0, score)


def _company_scale_score(candidate: dict) -> float:
    p = profile(candidate)
    role = current_role(candidate)
    score = 0.0
    for size in (p.get("current_company_size"), role.get("company_size")):
        if str(size or "") in STARTUP_SIZES:
            score += 0.35
        elif str(size or "") in {"501-1000", "1001-5000"}:
            score += 0.20
        elif str(size or "") == "10001+":
            score += 0.05
    return min(1.0, score)


def _work_mode_fit(candidate: dict) -> float:
    mode = _lower((candidate.get("redrob_signals") or {}).get("preferred_work_mode"))
    if mode in ("hybrid", "flexible"):
        return 1.0
    if mode == "remote":
        return 0.82
    if mode == "onsite":
        return 0.90
    return 0.70


def _platform_activity_score(candidate: dict) -> float:
    sig = candidate.get("redrob_signals") or {}
    views = int(sig.get("profile_views_received_30d") or 0)
    apps = int(sig.get("applications_submitted_30d") or 0)
    search = int(sig.get("search_appearance_30d") or 0)
    saved = int(sig.get("saved_by_recruiters_30d") or 0)
    score = 0.0
    if views >= 20:
        score += 0.25
    elif views >= 5:
        score += 0.12
    if apps >= 3:
        score += 0.15
    if search >= 10:
        score += 0.25
    elif search >= 3:
        score += 0.12
    if saved >= 3:
        score += 0.20
    elif saved >= 1:
        score += 0.10
    return min(1.0, score)


def _verified_trust(candidate: dict) -> float:
    sig = candidate.get("redrob_signals") or {}
    flags = [bool(sig.get("verified_email")), bool(sig.get("verified_phone")), bool(sig.get("linkedin_connected"))]
    return sum(flags) / 3.0


def _yoe_location_fit(candidate: dict, spec: dict) -> float:
    p = profile(candidate)
    sig = candidate.get("redrob_signals") or {}
    yoe = float(p.get("years_of_experience") or 0)
    yspec = spec.get("yoe", {})
    ymin, ymax = float(yspec.get("min", 5.0)), float(yspec.get("max", 9.0))
    if ymin <= yoe <= ymax:
        yoe_score = 1.0
    elif yoe < ymin:
        yoe_score = max(0.25, 1.0 - (ymin - yoe) / 4.0)
    else:
        yoe_score = max(0.35, 1.0 - (yoe - ymax) / 7.0)
    loc_text = f"{p.get('location', '')} {p.get('country', '')}".lower()
    boosts = [_lower(x) for x in spec.get("location_boost", [])]
    loc_score = 1.0 if any(x and x in loc_text for x in boosts) else 0.35
    if sig.get("willing_to_relocate"):
        loc_score = max(loc_score, 0.70)
    return 0.60 * yoe_score + 0.40 * loc_score


def _consulting_only(candidate: dict, spec: dict) -> bool:
    roles = candidate.get("career_history") or []
    if len(roles) < 2:
        return False
    firms = tuple(_lower(x) for x in spec.get("consulting_companies", DEFAULT_CONSULTING))
    consulting_roles = 0
    for role in roles:
        blob = f"{role.get('company','')} {role.get('industry','')} {role.get('description','')}".lower()
        if any(firm in blob for firm in firms) or any(term in blob for term in ("consulting", "outsourc", "staffing")):
            consulting_roles += 1
    return consulting_roles == len(roles)


def _product_company_score(candidate: dict, spec: dict) -> float:
    roles = candidate.get("career_history") or []
    if not roles:
        return 0.0
    terms = tuple(_lower(x) for x in spec.get("product_company_terms", []))
    score = 0.0
    for role in roles:
        blob = f"{role.get('company','')} {role.get('industry','')} {role.get('description','')}".lower()
        if any(term in blob for term in terms):
            score += 0.35
        if not any(firm in blob for firm in DEFAULT_CONSULTING):
            score += 0.10
    return min(1.0, score)


def _research_only(candidate: dict, career_ev: float) -> bool:
    blob = full_text(candidate).lower()
    research = any(term in blob for term in ("research scientist", "academic lab", "phd", "paper", "publication", "university lab"))
    production = any(term in blob for term in ("production", "shipped", "deployed", "real users", "a/b test", "serving"))
    return research and not production and career_ev < 0.35


def _langchain_only(candidate: dict, career_ev: float) -> bool:
    blob = full_text(candidate).lower()
    has_framework = "langchain" in blob or "llamaindex" in blob
    real_ir = any(term in blob for term in ("retrieval", "ranking", "faiss", "bm25", "elasticsearch", "recommendation", "ndcg", "mrr"))
    return has_framework and not real_ir and career_ev < 0.35


def _title_chaser(candidate: dict) -> bool:
    roles = sorted(candidate.get("career_history") or [], key=lambda r: str(r.get("start_date") or ""), reverse=True)
    recent = roles[:5]
    short_roles = sum(1 for role in recent if int(role.get("duration_months") or 0) < 18)
    title_words = " ".join(_lower(r.get("title")) for r in recent)
    senior_titles = sum(1 for word in ("senior", "staff", "principal", "lead") if word in title_words)
    return len(recent) >= 4 and short_roles >= 4 and senior_titles >= 2


def _summary_title_mismatch(candidate: dict) -> bool:
    p = profile(candidate)
    summary = _lower(p.get("summary"))
    title = _lower(p.get("current_title"))
    if "marketing manager" in summary and "marketing" not in title:
        return True
    if "my professional background is in marketing manager" in summary and "marketing" not in title:
        return True
    return False


def _cv_speech_without_ir(candidate: dict) -> bool:
    blob = full_text(candidate).lower()
    cv_hits = sum(1 for term in CV_SPEECH_ROBOTICS if term in blob)
    ir_hits = sum(1 for term in ("retrieval", "ranking", "recommendation", "nlp", "search", "semantic") if term in blob)
    return cv_hits >= 3 and ir_hits == 0


def honeypot_flag(candidate: dict, title_tier: str, career_ev: float, ai_count: int, trust: float) -> tuple[bool, list[str]]:
    p = profile(candidate)
    yoe = float(p.get("years_of_experience") or 0)
    skills = candidate.get("skills") or []
    roles = candidate.get("career_history") or []
    reasons: list[str] = []

    expert_tiny = [s for s in skills if _lower(s.get("proficiency")) == "expert" and float(s.get("duration_months") or 0) < 3]
    if len(expert_tiny) >= 3:
        reasons.append("expert skills with near-zero duration")
    expert_zero = [s for s in skills if _lower(s.get("proficiency")) == "expert" and float(s.get("duration_months") or 0) == 0]
    if expert_zero:
        reasons.append("expert skill with zero duration")
    if len(skills) >= 10 and sum(int(s.get("endorsements") or 0) for s in skills) == 0:
        reasons.append("many skills with zero endorsements")

    years = [_year(r.get("start_date")) for r in roles if _year(r.get("start_date"))]
    if years and yoe >= 8 and min(years) > datetime.now().year - 4:
        reasons.append("claimed YoE contradicts role start dates")

    total_role_months = sum(int(r.get("duration_months") or 0) for r in roles)
    if yoe > 0 and total_role_months > int(yoe * 12 + 30):
        reasons.append("role durations exceed total experience")
    if yoe > 0 and any(int(r.get("duration_months") or 0) > int(yoe * 12 + 6) for r in roles):
        reasons.append("single role longer than total experience")

    edu_end_years = [int(e.get("end_year")) for e in candidate.get("education") or [] if e.get("end_year")]
    if edu_end_years and years:
        earliest_job = min(years)
        latest_edu = max(edu_end_years)
        if latest_edu > earliest_job + 2 and yoe >= 6:
            reasons.append("education timeline conflicts with claimed experience")

    assess = assessment_score(candidate)
    if assess >= 0.85 and title_tier == "trap" and career_ev < 0.20:
        reasons.append("high assessments but no matching career proof")
    if title_tier in ("trap", "weak") and ai_count >= 8 and trust < 0.20 and career_ev < 0.30:
        reasons.append("keyword-stuffed profile without career proof")
    return bool(reasons), reasons[:4]


def extract_features(candidate: dict, spec: dict) -> dict:
    p = profile(candidate)
    sig = candidate.get("redrob_signals") or {}
    role = current_role(candidate)
    evidence, evidence_terms, career_only_ev, summary_ev = _blended_evidence(candidate, spec)
    title = str(p.get("current_title") or "")
    tier, tier_score, best_title = _best_title_tier(candidate, spec)
    trust, ai_count, ai_skill_names = skill_trust(candidate)
    assess = assessment_score(candidate)
    consulting = _consulting_only(candidate, spec)
    stuffer = tier in ("weak", "trap") and ai_count >= 6 and trust < 0.35 and evidence < 0.45
    research_only = _research_only(candidate, evidence)
    langchain_only = _langchain_only(candidate, evidence)
    title_chaser = _title_chaser(candidate)
    mismatch = _summary_title_mismatch(candidate)
    cv_only = _cv_speech_without_ir(candidate)
    hp, hp_reasons = honeypot_flag(candidate, tier, evidence, ai_count, trust)

    anti = 0.0
    if stuffer:
        anti += 0.50
    if consulting:
        anti += 0.30
    if research_only:
        anti += 0.40
    if langchain_only:
        anti += 0.25
    if title_chaser:
        anti += 0.20
    if mismatch:
        anti += 0.35
    if cv_only:
        anti += 0.20
    if tier == "trap" and evidence < 0.60:
        anti += 0.80

    ctext = career_text(candidate)
    edu_top = ""
    if candidate.get("education"):
        e0 = candidate["education"][0]
        edu_top = f"{e0.get('degree','')} {e0.get('field_of_study','')}".strip()

    return {
        "candidate_id": candidate.get("candidate_id"),
        "current_title": title,
        "best_career_title": best_title,
        "current_company": str(p.get("current_company") or ""),
        "current_industry": str(p.get("current_industry") or ""),
        "years_of_experience": float(p.get("years_of_experience") or 0),
        "location": str(p.get("location") or ""),
        "country": str(p.get("country") or ""),
        "career_evidence": evidence,
        "career_only_evidence": career_only_ev,
        "summary_evidence": summary_ev,
        "career_terms": "|".join(evidence_terms),
        "title_tier": tier,
        "title_tier_score": tier_score,
        "skill_trust": min(1.0, trust),
        "ai_skill_count": ai_count,
        "ai_skill_names": "|".join(ai_skill_names[:5]),
        "assessment_score": assess,
        "education_score": _education_score(candidate),
        "education_top": edu_top,
        "company_scale_score": _company_scale_score(candidate),
        "work_mode_fit": _work_mode_fit(candidate),
        "platform_activity_score": _platform_activity_score(candidate),
        "verified_trust": _verified_trust(candidate),
        "yoe_location_fit": _yoe_location_fit(candidate, spec),
        "product_company_score": _product_company_score(candidate, spec),
        "anti_pattern_penalty": min(1.0, anti),
        "honeypot_flag": hp,
        "honeypot_reasons": "|".join(hp_reasons),
        "stuffer_flag": stuffer,
        "consulting_only": consulting,
        "research_only": research_only,
        "langchain_only": langchain_only,
        "title_chaser": title_chaser,
        "summary_title_mismatch": mismatch,
        "cv_speech_without_ir": cv_only,
        "preferred_work_mode": str(sig.get("preferred_work_mode") or ""),
        "open_to_work_flag": bool(sig.get("open_to_work_flag", False)),
        "willing_to_relocate": bool(sig.get("willing_to_relocate", False)),
        "recruiter_response_rate": float(sig.get("recruiter_response_rate") or 0),
        "avg_response_time_hours": float(sig.get("avg_response_time_hours") or 0),
        "notice_period_days": int(sig.get("notice_period_days") or 90),
        "profile_completeness_score": float(sig.get("profile_completeness_score") or 0),
        "last_active_date": sig.get("last_active_date") or "",
        "github_activity_score": float(sig.get("github_activity_score") if sig.get("github_activity_score") is not None else -1),
        "saved_by_recruiters_30d": int(sig.get("saved_by_recruiters_30d") or 0),
        "profile_views_received_30d": int(sig.get("profile_views_received_30d") or 0),
        "search_appearance_30d": int(sig.get("search_appearance_30d") or 0),
        "interview_completion_rate": float(sig.get("interview_completion_rate") or 0),
        "offer_acceptance_rate": float(sig.get("offer_acceptance_rate") if sig.get("offer_acceptance_rate") is not None else -1),
        "verified_email": bool(sig.get("verified_email", False)),
        "verified_phone": bool(sig.get("verified_phone", False)),
        "linkedin_connected": bool(sig.get("linkedin_connected", False)),
        "career_hash": hashlib.sha1(ctext.encode("utf-8", errors="ignore")).hexdigest(),
        "career_hash_count": 1,
        "current_role_excerpt": str(role.get("description") or "")[:180].replace("\n", " "),
    }


def attach_hash_counts(features_df):
    counts = features_df["career_hash"].value_counts().to_dict()
    features_df["career_hash_count"] = features_df["career_hash"].map(counts).astype(int)
    return features_df


def features_dataframe(candidates: list[dict], spec: dict):
    import pandas as pd

    df = pd.DataFrame(extract_features(candidate, spec) for candidate in candidates)
    return attach_hash_counts(df)