Spaces:
Sleeping
Sleeping
Update app/main.py
Browse files- app/main.py +183 -721
app/main.py
CHANGED
|
@@ -1,19 +1,17 @@
|
|
|
|
|
| 1 |
import json
|
| 2 |
import os
|
| 3 |
import re
|
| 4 |
-
import tempfile
|
| 5 |
-
from pathlib import Path
|
| 6 |
from statistics import mean
|
| 7 |
-
from typing import Any, Dict, List
|
| 8 |
|
| 9 |
import requests
|
| 10 |
from fastapi import FastAPI, File, Form, UploadFile
|
| 11 |
from fastapi.middleware.cors import CORSMiddleware
|
| 12 |
from pydantic import BaseModel
|
| 13 |
|
| 14 |
-
|
| 15 |
|
| 16 |
-
app = FastAPI(title="Japanese AI Interview API", version=APP_VERSION)
|
| 17 |
app.add_middleware(
|
| 18 |
CORSMiddleware,
|
| 19 |
allow_origins=["*"],
|
|
@@ -24,110 +22,150 @@ app.add_middleware(
|
|
| 24 |
|
| 25 |
HF_TOKEN = os.getenv("HF_TOKEN", "").strip()
|
| 26 |
ASR_MODEL = os.getenv("ASR_MODEL", "openai/whisper-large-v3")
|
| 27 |
-
CHAT_MODEL = os.getenv("CHAT_MODEL", "Qwen/Qwen2.5-7B-Instruct-1M")
|
| 28 |
-
HF_ROUTER_URL = os.getenv("HF_ROUTER_URL", "https://router.huggingface.co/v1/chat/completions")
|
| 29 |
HF_INFERENCE_BASE = os.getenv("HF_INFERENCE_BASE", "https://router.huggingface.co/hf-inference/models")
|
| 30 |
-
MAX_QUESTION_LIMIT = int(os.getenv("MAX_QUESTION_LIMIT", "20"))
|
| 31 |
-
LLM_TIMEOUT_SECONDS = int(os.getenv("LLM_TIMEOUT_SECONDS", "90"))
|
| 32 |
-
ASR_TIMEOUT_SECONDS = int(os.getenv("ASR_TIMEOUT_SECONDS", "180"))
|
| 33 |
|
| 34 |
-
|
| 35 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 36 |
|
| 37 |
-
REPEAT_PROMPTS = [
|
| 38 |
-
"ใใฟใพใใใ้ณใ่ใใใพใใใงใใใใใคใฏใ็ขบ่ชใใฆใใใไธๅบฆใ้กใใใพใใ",
|
| 39 |
-
"ๅฃฐใๅฐใใใงใใใใๅฐใๅคงใใๅฃฐใงใใใไธๅบฆใ้กใใใพใใ",
|
| 40 |
-
"ใพใ ้ณใใใพใๅ
ฅใใพใใใใใคใฏใ่ฟใฅใใฆใใใไธๅบฆใ้กใใใพใใ",
|
| 41 |
-
]
|
| 42 |
|
| 43 |
class StartRequest(BaseModel):
|
| 44 |
session_uuid: str
|
| 45 |
job_role: str = "construction"
|
| 46 |
-
question_count: int = 10
|
| 47 |
|
| 48 |
|
| 49 |
@app.get("/")
|
| 50 |
def root() -> Dict[str, Any]:
|
| 51 |
-
return {
|
| 52 |
-
"ok": True,
|
| 53 |
-
"service": "jp-role-interview",
|
| 54 |
-
"version": APP_VERSION,
|
| 55 |
-
"routes": ["/health", "/roles", "/start", "/answer"],
|
| 56 |
-
}
|
| 57 |
|
| 58 |
|
| 59 |
@app.get("/health")
|
| 60 |
def health() -> Dict[str, Any]:
|
| 61 |
return {
|
| 62 |
"ok": True,
|
| 63 |
-
"service": "jp-
|
| 64 |
-
"version":
|
| 65 |
"hf_token_set": bool(HF_TOKEN),
|
| 66 |
"asr_model": ASR_MODEL,
|
| 67 |
-
"
|
| 68 |
-
"role_count": len(ROLE_BANK),
|
| 69 |
-
"native_asr": False,
|
| 70 |
-
"uses_hf_serverless_asr": True,
|
| 71 |
}
|
| 72 |
|
| 73 |
|
| 74 |
@app.get("/roles")
|
| 75 |
def roles() -> Dict[str, Any]:
|
| 76 |
-
|
| 77 |
-
|
| 78 |
-
|
| 79 |
-
|
| 80 |
-
"english_name": role["english_name"],
|
| 81 |
-
"japanese_name": role["japanese_name"],
|
| 82 |
-
"question_count": role["question_count"],
|
| 83 |
-
"min_questions": role["min_questions"],
|
| 84 |
-
"max_questions": role["max_questions"],
|
| 85 |
-
})
|
| 86 |
-
return {"ok": True, "roles": items}
|
| 87 |
|
| 88 |
|
| 89 |
@app.post("/start")
|
| 90 |
def start_interview(payload: StartRequest) -> Dict[str, Any]:
|
| 91 |
role_key = payload.job_role if payload.job_role in ROLE_BANK else "construction"
|
| 92 |
-
|
| 93 |
-
question_count = max(role["min_questions"], min(payload.question_count, role["max_questions"], MAX_QUESTION_LIMIT))
|
| 94 |
-
|
| 95 |
-
opening_question = f"{role['intro_jp']} ใพใใใๅๅใๆใใฆใใ ใใใ"
|
| 96 |
-
first_question = get_question_by_id(role, f"{role_key}_common_name_1")
|
| 97 |
-
if first_question:
|
| 98 |
-
opening_question = f"{role['intro_jp']} {first_question['jp']}"
|
| 99 |
-
|
| 100 |
memory = {
|
| 101 |
-
"session_uuid": payload.session_uuid,
|
| 102 |
"job_role": role_key,
|
| 103 |
-
"
|
| 104 |
-
"job_role_jp": role["japanese_name"],
|
| 105 |
-
"question_count_target": question_count,
|
| 106 |
"candidate_name": None,
|
| 107 |
-
"country_name": None,
|
| 108 |
-
"age": None,
|
| 109 |
-
"reason_for_japan": None,
|
| 110 |
-
"occupation": None,
|
| 111 |
-
"japanese_level": None,
|
| 112 |
-
"experience_status": "unknown",
|
| 113 |
"answers_so_far": [],
|
| 114 |
-
"asked_question_ids": [first_question["id"]] if first_question else [],
|
| 115 |
-
"asked_themes": ["intro"],
|
| 116 |
"low_score_count": 0,
|
| 117 |
"no_sound_count": 0,
|
| 118 |
-
"
|
| 119 |
-
"question_pool_size": role["question_count"],
|
| 120 |
-
"interview_status": "running",
|
| 121 |
}
|
| 122 |
-
|
| 123 |
return {
|
| 124 |
"ok": True,
|
| 125 |
"session_uuid": payload.session_uuid,
|
| 126 |
"job_role": role_key,
|
|
|
|
| 127 |
"question_no": 1,
|
| 128 |
-
"
|
| 129 |
-
"
|
| 130 |
-
"question_jp": opening_question,
|
| 131 |
"memory": memory,
|
| 132 |
"is_finished": False,
|
| 133 |
"speak_now": True,
|
|
@@ -138,187 +176,139 @@ def start_interview(payload: StartRequest) -> Dict[str, Any]:
|
|
| 138 |
async def answer_interview(
|
| 139 |
session_uuid: str = Form(...),
|
| 140 |
question_no: int = Form(...),
|
| 141 |
-
question_count: int = Form(10),
|
| 142 |
-
question_id: str = Form(""),
|
| 143 |
question_jp: str = Form(...),
|
| 144 |
memory_json: str = Form("{}"),
|
| 145 |
audio: UploadFile = File(...),
|
| 146 |
) -> Dict[str, Any]:
|
| 147 |
memory = safe_json_loads(memory_json)
|
| 148 |
-
role_key = memory.get("job_role")
|
| 149 |
-
|
| 150 |
-
question_count = max(role["min_questions"], min(question_count, role["max_questions"], MAX_QUESTION_LIMIT))
|
| 151 |
|
| 152 |
audio_bytes = await audio.read()
|
| 153 |
transcript = ""
|
| 154 |
-
|
| 155 |
-
if audio_bytes:
|
| 156 |
try:
|
| 157 |
transcript = transcribe_audio_with_hf(audio_bytes, audio.filename or "audio.webm")
|
| 158 |
-
except Exception
|
| 159 |
-
|
| 160 |
|
| 161 |
if not transcript.strip():
|
| 162 |
memory["no_sound_count"] = int(memory.get("no_sound_count", 0)) + 1
|
| 163 |
-
|
| 164 |
-
|
| 165 |
-
|
| 166 |
-
if memory["no_sound_count"] >= 2 and question_no >= role["min_questions"]:
|
| 167 |
-
result = build_final_result(memory, role, force_fail=True, summary_jp="้ณๅฃฐใ่ใใใชใใใใ้ขๆฅใ็ตไบใใพใใใ")
|
| 168 |
-
return {
|
| 169 |
-
"ok": True,
|
| 170 |
-
"is_finished": True,
|
| 171 |
-
"session_uuid": session_uuid,
|
| 172 |
-
"question_no": question_no,
|
| 173 |
-
"question_count": question_count,
|
| 174 |
-
"transcript_jp": "",
|
| 175 |
-
"answer_score": 0,
|
| 176 |
-
"feedback_jp": repeat_prompt,
|
| 177 |
-
"memory": memory,
|
| 178 |
-
"llm_used": False,
|
| 179 |
-
"result": result,
|
| 180 |
-
}
|
| 181 |
return {
|
| 182 |
"ok": True,
|
| 183 |
"is_finished": False,
|
| 184 |
"needs_repeat": True,
|
| 185 |
"session_uuid": session_uuid,
|
| 186 |
"question_no": question_no,
|
| 187 |
-
"question_count": question_count,
|
| 188 |
-
"question_id": question_id,
|
| 189 |
"question_jp": question_jp,
|
|
|
|
| 190 |
"transcript_jp": "",
|
| 191 |
"answer_score": 0,
|
| 192 |
-
"feedback_jp":
|
| 193 |
"memory": memory,
|
| 194 |
"next_question_no": question_no,
|
| 195 |
-
"next_question_id": question_id,
|
| 196 |
"next_question_jp": question_jp,
|
| 197 |
"speak_now": True,
|
| 198 |
-
"asr_error": asr_error,
|
| 199 |
-
"llm_used": False,
|
| 200 |
}
|
| 201 |
|
| 202 |
-
|
|
|
|
|
|
|
|
|
|
| 203 |
|
| 204 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 205 |
"question_no": question_no,
|
| 206 |
-
"question_id": question_id,
|
| 207 |
"question_jp": question_jp,
|
| 208 |
"answer_text_jp": transcript,
|
| 209 |
-
|
| 210 |
-
|
| 211 |
-
history
|
| 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 |
-
merged_memory["asked_question_ids"] = merge_unique(memory.get("asked_question_ids", []), [evaluation["next_question_id"]])
|
| 241 |
-
|
| 242 |
-
answer_score = clamp_int(evaluation.get("answer_score", heuristic_score(transcript, role)), 0, 10)
|
| 243 |
-
feedback_jp = clean_text(evaluation.get("feedback_jp")) or default_feedback(answer_score)
|
| 244 |
-
|
| 245 |
-
history[-1]["answer_score"] = answer_score
|
| 246 |
-
history[-1]["feedback_jp"] = feedback_jp
|
| 247 |
-
history[-1]["question_theme"] = evaluation.get("question_theme")
|
| 248 |
-
history[-1]["keywords_matched"] = keyword_matches(transcript, role["expected_keywords"])
|
| 249 |
-
|
| 250 |
-
merged_memory["experience_status"] = detect_experience_status(merged_memory, transcript)
|
| 251 |
-
merged_memory["low_score_count"] = int(memory.get("low_score_count", 0)) + (1 if answer_score <= 3 else 0)
|
| 252 |
-
|
| 253 |
-
should_finish = decide_finish(
|
| 254 |
-
role=role,
|
| 255 |
-
memory=merged_memory,
|
| 256 |
-
question_no=question_no,
|
| 257 |
-
question_count=question_count,
|
| 258 |
-
answer_score=answer_score,
|
| 259 |
-
)
|
| 260 |
-
|
| 261 |
-
if should_finish:
|
| 262 |
-
result = run_final_evaluation_if_possible(merged_memory, role, llm_used=llm_used)
|
| 263 |
return {
|
| 264 |
"ok": True,
|
| 265 |
"is_finished": True,
|
| 266 |
"session_uuid": session_uuid,
|
| 267 |
"question_no": question_no,
|
| 268 |
-
"question_count": question_count,
|
| 269 |
"transcript_jp": transcript,
|
| 270 |
-
"answer_score":
|
| 271 |
-
"feedback_jp":
|
| 272 |
-
"
|
| 273 |
-
"
|
| 274 |
"result": result,
|
| 275 |
}
|
| 276 |
|
| 277 |
next_question_no = question_no + 1
|
| 278 |
-
|
| 279 |
-
|
| 280 |
-
|
| 281 |
-
|
| 282 |
-
chosen = candidate_questions[0] if candidate_questions else choose_any_unused_question(role, merged_memory)
|
| 283 |
-
next_question_id = chosen["id"]
|
| 284 |
-
next_question_jp = chosen["jp"]
|
| 285 |
|
| 286 |
return {
|
| 287 |
"ok": True,
|
| 288 |
"is_finished": False,
|
| 289 |
"session_uuid": session_uuid,
|
| 290 |
"question_no": question_no,
|
| 291 |
-
"question_count": question_count,
|
| 292 |
"transcript_jp": transcript,
|
| 293 |
-
"answer_score":
|
| 294 |
-
"feedback_jp":
|
| 295 |
-
"
|
| 296 |
-
"
|
| 297 |
"next_question_no": next_question_no,
|
| 298 |
-
"
|
| 299 |
-
"next_question_jp": next_question_jp,
|
| 300 |
"speak_now": True,
|
| 301 |
}
|
| 302 |
|
| 303 |
|
| 304 |
def transcribe_audio_with_hf(audio_bytes: bytes, filename: str) -> str:
|
| 305 |
-
if not HF_TOKEN:
|
| 306 |
-
raise RuntimeError("HF_TOKEN is missing for ASR.")
|
| 307 |
url = f"{HF_INFERENCE_BASE}/{ASR_MODEL}"
|
| 308 |
headers = {
|
| 309 |
"Authorization": f"Bearer {HF_TOKEN}",
|
| 310 |
"Content-Type": guess_mime_type(filename),
|
| 311 |
}
|
| 312 |
-
response = requests.post(url, headers=headers, data=audio_bytes, timeout=
|
| 313 |
response.raise_for_status()
|
| 314 |
data = response.json()
|
| 315 |
if isinstance(data, dict):
|
| 316 |
-
|
| 317 |
-
|
| 318 |
-
|
| 319 |
-
# fallback for some provider formats
|
| 320 |
-
text = data[0].get("text", "") if isinstance(data[0], dict) else ""
|
| 321 |
-
return normalize_text(text)
|
| 322 |
return ""
|
| 323 |
|
| 324 |
|
|
@@ -335,553 +325,36 @@ def guess_mime_type(filename: str) -> str:
|
|
| 335 |
return "audio/webm"
|
| 336 |
|
| 337 |
|
| 338 |
-
def run_llm_turn(
|
| 339 |
-
role: Dict[str, Any],
|
| 340 |
-
question_no: int,
|
| 341 |
-
question_count: int,
|
| 342 |
-
question_id: str,
|
| 343 |
-
current_question: str,
|
| 344 |
-
transcript: str,
|
| 345 |
-
memory: Dict[str, Any],
|
| 346 |
-
history: List[Dict[str, Any]],
|
| 347 |
-
candidate_questions: List[Dict[str, Any]],
|
| 348 |
-
) -> Dict[str, Any]:
|
| 349 |
-
system_prompt = (
|
| 350 |
-
"You are a Japanese interviewer for working visa practice. "
|
| 351 |
-
"Use simple N4-level Japanese. "
|
| 352 |
-
"Ask ONE realistic interview question at a time. "
|
| 353 |
-
"Stay inside the selected job role. "
|
| 354 |
-
"If the answer is weak or unclear, you may ask a short repeat or clarification question. "
|
| 355 |
-
"If the candidate is failing badly and the minimum number of questions has been reached, you may end the interview early. "
|
| 356 |
-
"Return ONLY valid JSON."
|
| 357 |
-
)
|
| 358 |
-
payload = {
|
| 359 |
-
"role_key": role["role_key"],
|
| 360 |
-
"role_name_jp": role["japanese_name"],
|
| 361 |
-
"question_no": question_no,
|
| 362 |
-
"question_count_target": question_count,
|
| 363 |
-
"current_question_id": question_id,
|
| 364 |
-
"current_question_jp": current_question,
|
| 365 |
-
"candidate_answer_jp": transcript,
|
| 366 |
-
"memory": {
|
| 367 |
-
"candidate_name": memory.get("candidate_name"),
|
| 368 |
-
"country_name": memory.get("country_name"),
|
| 369 |
-
"age": memory.get("age"),
|
| 370 |
-
"reason_for_japan": memory.get("reason_for_japan"),
|
| 371 |
-
"occupation": memory.get("occupation"),
|
| 372 |
-
"japanese_level": memory.get("japanese_level"),
|
| 373 |
-
"experience_status": memory.get("experience_status"),
|
| 374 |
-
"low_score_count": memory.get("low_score_count", 0),
|
| 375 |
-
"asked_themes": memory.get("asked_themes", []),
|
| 376 |
-
},
|
| 377 |
-
"history_tail": history[-6:],
|
| 378 |
-
"candidate_questions": [
|
| 379 |
-
{
|
| 380 |
-
"id": q["id"],
|
| 381 |
-
"jp": q["jp"],
|
| 382 |
-
"theme": q["theme"],
|
| 383 |
-
"branch": q["branch"],
|
| 384 |
-
"stage": q["stage"],
|
| 385 |
-
"expected_keywords": q.get("expected_keywords", []),
|
| 386 |
-
}
|
| 387 |
-
for q in candidate_questions[:12]
|
| 388 |
-
],
|
| 389 |
-
"schema": {
|
| 390 |
-
"answer_score": "integer 0-10",
|
| 391 |
-
"feedback_jp": "one short Japanese sentence",
|
| 392 |
-
"question_theme": "theme string",
|
| 393 |
-
"asked_themes": ["array"],
|
| 394 |
-
"profile_update": {
|
| 395 |
-
"candidate_name": "string or null",
|
| 396 |
-
"country_name": "string or null",
|
| 397 |
-
"age": "integer or null",
|
| 398 |
-
"reason_for_japan": "string or null",
|
| 399 |
-
"occupation": "string or null",
|
| 400 |
-
"japanese_level": "string or null",
|
| 401 |
-
"experience_status": "yes or no or unknown"
|
| 402 |
-
},
|
| 403 |
-
"continue_interview": "boolean",
|
| 404 |
-
"next_question_id": "question id from candidate_questions or empty when ending",
|
| 405 |
-
"next_question_jp": "question text from candidate_questions or empty when ending"
|
| 406 |
-
},
|
| 407 |
-
"rules": [
|
| 408 |
-
"Use only the provided candidate_questions for next_question_id.",
|
| 409 |
-
"Do not repeat a question unless clarification is needed.",
|
| 410 |
-
"Keep the question natural and interview-like.",
|
| 411 |
-
"Use the candidate name if known.",
|
| 412 |
-
"If question_no is already at target, continue_interview must be false."
|
| 413 |
-
],
|
| 414 |
-
}
|
| 415 |
-
raw = call_hf_chat_json(system_prompt=system_prompt, user_payload=payload)
|
| 416 |
-
result = normalize_llm_turn_result(raw)
|
| 417 |
-
result["profile_update"] = merge_memory(
|
| 418 |
-
maybe_extract_basic_profile(memory, transcript, question_no),
|
| 419 |
-
result.get("profile_update", {}),
|
| 420 |
-
)
|
| 421 |
-
if question_no >= question_count:
|
| 422 |
-
result["continue_interview"] = False
|
| 423 |
-
result["next_question_id"] = ""
|
| 424 |
-
result["next_question_jp"] = ""
|
| 425 |
-
return result
|
| 426 |
-
|
| 427 |
-
|
| 428 |
-
def normalize_llm_turn_result(raw: Dict[str, Any]) -> Dict[str, Any]:
|
| 429 |
-
profile = raw.get("profile_update", {}) if isinstance(raw.get("profile_update"), dict) else {}
|
| 430 |
-
asked_themes = raw.get("asked_themes", [])
|
| 431 |
-
if not isinstance(asked_themes, list):
|
| 432 |
-
asked_themes = []
|
| 433 |
-
return {
|
| 434 |
-
"answer_score": clamp_int(raw.get("answer_score", 6), 0, 10),
|
| 435 |
-
"feedback_jp": clean_text(raw.get("feedback_jp")),
|
| 436 |
-
"question_theme": clean_text(raw.get("question_theme")) or None,
|
| 437 |
-
"asked_themes": [clean_text(x) for x in asked_themes if clean_text(x)],
|
| 438 |
-
"profile_update": {
|
| 439 |
-
"candidate_name": normalize_optional_text(profile.get("candidate_name")),
|
| 440 |
-
"country_name": normalize_optional_text(profile.get("country_name")),
|
| 441 |
-
"age": normalize_optional_int(profile.get("age")),
|
| 442 |
-
"reason_for_japan": normalize_optional_text(profile.get("reason_for_japan")),
|
| 443 |
-
"occupation": normalize_optional_text(profile.get("occupation")),
|
| 444 |
-
"japanese_level": normalize_optional_text(profile.get("japanese_level")),
|
| 445 |
-
"experience_status": normalize_optional_text(profile.get("experience_status")),
|
| 446 |
-
},
|
| 447 |
-
"continue_interview": bool(raw.get("continue_interview", True)),
|
| 448 |
-
"next_question_id": clean_text(raw.get("next_question_id")),
|
| 449 |
-
"next_question_jp": clean_text(raw.get("next_question_jp")),
|
| 450 |
-
}
|
| 451 |
-
|
| 452 |
-
|
| 453 |
-
def call_hf_chat_json(system_prompt: str, user_payload: Dict[str, Any]) -> Dict[str, Any]:
|
| 454 |
-
if not HF_TOKEN:
|
| 455 |
-
raise RuntimeError("HF_TOKEN is missing.")
|
| 456 |
-
body = {
|
| 457 |
-
"model": CHAT_MODEL,
|
| 458 |
-
"messages": [
|
| 459 |
-
{"role": "system", "content": system_prompt},
|
| 460 |
-
{"role": "user", "content": json.dumps(user_payload, ensure_ascii=False)},
|
| 461 |
-
],
|
| 462 |
-
"temperature": 0.35,
|
| 463 |
-
"max_tokens": 900,
|
| 464 |
-
"response_format": {"type": "json_object"},
|
| 465 |
-
}
|
| 466 |
-
response = requests.post(
|
| 467 |
-
HF_ROUTER_URL,
|
| 468 |
-
headers={
|
| 469 |
-
"Authorization": f"Bearer {HF_TOKEN}",
|
| 470 |
-
"Content-Type": "application/json",
|
| 471 |
-
},
|
| 472 |
-
json=body,
|
| 473 |
-
timeout=LLM_TIMEOUT_SECONDS,
|
| 474 |
-
)
|
| 475 |
-
response.raise_for_status()
|
| 476 |
-
data = response.json()
|
| 477 |
-
content = data.get("choices", [{}])[0].get("message", {}).get("content", "")
|
| 478 |
-
if not content:
|
| 479 |
-
raise RuntimeError("HF router returned empty content.")
|
| 480 |
-
parsed = json.loads(content)
|
| 481 |
-
if not isinstance(parsed, dict):
|
| 482 |
-
raise RuntimeError("HF router did not return a JSON object.")
|
| 483 |
-
return parsed
|
| 484 |
-
|
| 485 |
-
|
| 486 |
-
def select_candidate_questions(role: Dict[str, Any], memory: Dict[str, Any], transcript: str) -> List[Dict[str, Any]]:
|
| 487 |
-
questions = role["questions"]
|
| 488 |
-
asked_ids = set(memory.get("asked_question_ids", []))
|
| 489 |
-
asked_themes = set(memory.get("asked_themes", []))
|
| 490 |
-
experience_status = detect_experience_status(memory, transcript)
|
| 491 |
-
priority: List[Tuple[int, Dict[str, Any]]] = []
|
| 492 |
-
|
| 493 |
-
for q in questions:
|
| 494 |
-
if q["id"] in asked_ids or q["stage"] == "closing":
|
| 495 |
-
continue
|
| 496 |
-
score = 0
|
| 497 |
-
theme = q["theme"]
|
| 498 |
-
stage = q["stage"]
|
| 499 |
-
branch = q["branch"]
|
| 500 |
-
|
| 501 |
-
if stage == "screening":
|
| 502 |
-
score += 30
|
| 503 |
-
if stage == "role":
|
| 504 |
-
score += 20
|
| 505 |
-
if stage == "followup":
|
| 506 |
-
score += 10
|
| 507 |
-
|
| 508 |
-
if theme not in asked_themes:
|
| 509 |
-
score += 15
|
| 510 |
-
else:
|
| 511 |
-
score -= 6
|
| 512 |
-
|
| 513 |
-
if branch == "yes_exp" and experience_status == "yes":
|
| 514 |
-
score += 18
|
| 515 |
-
elif branch == "no_exp" and experience_status == "no":
|
| 516 |
-
score += 18
|
| 517 |
-
elif branch in {"yes_exp", "no_exp"} and experience_status == "unknown":
|
| 518 |
-
score -= 10
|
| 519 |
-
|
| 520 |
-
if theme == "experience" and experience_status == "unknown":
|
| 521 |
-
score += 14
|
| 522 |
-
|
| 523 |
-
if theme in {"intro", "motivation", "language"} and len(memory.get("answers_so_far", [])) < 4:
|
| 524 |
-
score += 18
|
| 525 |
-
|
| 526 |
-
if theme in {"safety", "teamwork", "reliability"} and len(memory.get("answers_so_far", [])) >= 4:
|
| 527 |
-
score += 10
|
| 528 |
-
|
| 529 |
-
if keyword_matches(transcript, q.get("expected_keywords", [])):
|
| 530 |
-
score += 4
|
| 531 |
-
|
| 532 |
-
priority.append((score, q))
|
| 533 |
-
|
| 534 |
-
priority.sort(key=lambda x: x[0], reverse=True)
|
| 535 |
-
picked = [q for _, q in priority[:12]]
|
| 536 |
-
if not picked:
|
| 537 |
-
chosen = choose_any_unused_question(role, memory)
|
| 538 |
-
return [chosen] if chosen else []
|
| 539 |
-
return picked
|
| 540 |
-
|
| 541 |
-
|
| 542 |
-
def fallback_turn_evaluation(
|
| 543 |
-
role: Dict[str, Any],
|
| 544 |
-
memory: Dict[str, Any],
|
| 545 |
-
transcript: str,
|
| 546 |
-
candidate_questions: List[Dict[str, Any]],
|
| 547 |
-
error: str,
|
| 548 |
-
) -> Dict[str, Any]:
|
| 549 |
-
profile_update = maybe_extract_basic_profile(memory, transcript, len(memory.get("answers_so_far", [])))
|
| 550 |
-
if "experience_status" not in profile_update:
|
| 551 |
-
profile_update["experience_status"] = detect_experience_status(memory, transcript)
|
| 552 |
-
|
| 553 |
-
score = heuristic_score(transcript, role)
|
| 554 |
-
next_q = candidate_questions[0] if candidate_questions else choose_any_unused_question(role, memory)
|
| 555 |
-
continue_interview = True
|
| 556 |
-
if int(memory.get("low_score_count", 0)) >= 2 and len(memory.get("answers_so_far", [])) >= role["min_questions"]:
|
| 557 |
-
continue_interview = False
|
| 558 |
-
|
| 559 |
-
return {
|
| 560 |
-
"answer_score": score,
|
| 561 |
-
"feedback_jp": default_feedback(score),
|
| 562 |
-
"profile_update": profile_update,
|
| 563 |
-
"continue_interview": continue_interview,
|
| 564 |
-
"next_question_id": next_q["id"] if next_q else "",
|
| 565 |
-
"next_question_jp": next_q["jp"] if next_q else "",
|
| 566 |
-
"question_theme": next_q["theme"] if next_q else "",
|
| 567 |
-
"asked_themes": [next_q["theme"]] if next_q else [],
|
| 568 |
-
"debug_error": error,
|
| 569 |
-
}
|
| 570 |
-
|
| 571 |
-
|
| 572 |
-
def decide_finish(role: Dict[str, Any], memory: Dict[str, Any], question_no: int, question_count: int, answer_score: int) -> bool:
|
| 573 |
-
if question_no >= question_count:
|
| 574 |
-
return True
|
| 575 |
-
if int(memory.get("no_sound_count", 0)) >= 2 and question_no >= role["min_questions"]:
|
| 576 |
-
return True
|
| 577 |
-
if int(memory.get("low_score_count", 0)) >= 3 and question_no >= role["min_questions"]:
|
| 578 |
-
return True
|
| 579 |
-
return False
|
| 580 |
-
|
| 581 |
-
|
| 582 |
-
def run_final_evaluation_if_possible(merged_memory: Dict[str, Any], role: Dict[str, Any], llm_used: bool) -> Dict[str, Any]:
|
| 583 |
-
if llm_used and HF_TOKEN:
|
| 584 |
-
try:
|
| 585 |
-
return run_llm_final_evaluation(merged_memory, role)
|
| 586 |
-
except Exception:
|
| 587 |
-
pass
|
| 588 |
-
return build_final_result(merged_memory, role, force_fail=False)
|
| 589 |
-
|
| 590 |
-
|
| 591 |
-
def run_llm_final_evaluation(merged_memory: Dict[str, Any], role: Dict[str, Any]) -> Dict[str, Any]:
|
| 592 |
-
history = merged_memory.get("answers_so_far", [])
|
| 593 |
-
system_prompt = (
|
| 594 |
-
"You are a Japanese interview evaluator for N4-level job interview practice. "
|
| 595 |
-
"Review the interview fairly. "
|
| 596 |
-
"Return ONLY valid JSON. "
|
| 597 |
-
"Use short Japanese only for summary_jp and closing_message_jp."
|
| 598 |
-
)
|
| 599 |
-
payload = {
|
| 600 |
-
"role_key": role["role_key"],
|
| 601 |
-
"role_name_jp": role["japanese_name"],
|
| 602 |
-
"profile_memory": {
|
| 603 |
-
"candidate_name": merged_memory.get("candidate_name"),
|
| 604 |
-
"country_name": merged_memory.get("country_name"),
|
| 605 |
-
"age": merged_memory.get("age"),
|
| 606 |
-
"reason_for_japan": merged_memory.get("reason_for_japan"),
|
| 607 |
-
"occupation": merged_memory.get("occupation"),
|
| 608 |
-
"japanese_level": merged_memory.get("japanese_level"),
|
| 609 |
-
"experience_status": merged_memory.get("experience_status"),
|
| 610 |
-
},
|
| 611 |
-
"history": history,
|
| 612 |
-
"schema": {
|
| 613 |
-
"summary_jp": "short Japanese summary",
|
| 614 |
-
"overall_score": "integer 0-100",
|
| 615 |
-
"scores": {
|
| 616 |
-
"fluency": "1-10",
|
| 617 |
-
"grammar": "1-10",
|
| 618 |
-
"confidence": "1-10",
|
| 619 |
-
"relevance": "1-10",
|
| 620 |
-
"role_fit": "1-10"
|
| 621 |
-
},
|
| 622 |
-
"pass_fail": "PASS or FAIL",
|
| 623 |
-
"strengths": ["short strings"],
|
| 624 |
-
"weaknesses": ["short strings"],
|
| 625 |
-
"tips": ["short strings"],
|
| 626 |
-
"closing_message_jp": "thank you message in Japanese"
|
| 627 |
-
},
|
| 628 |
-
}
|
| 629 |
-
raw = call_hf_chat_json(system_prompt=system_prompt, user_payload=payload)
|
| 630 |
-
raw_scores = raw.get("scores", {}) if isinstance(raw.get("scores"), dict) else {}
|
| 631 |
-
return {
|
| 632 |
-
"candidate_name": merged_memory.get("candidate_name"),
|
| 633 |
-
"country_name": merged_memory.get("country_name"),
|
| 634 |
-
"age": merged_memory.get("age"),
|
| 635 |
-
"job_role": merged_memory.get("job_role"),
|
| 636 |
-
"job_role_jp": merged_memory.get("job_role_jp"),
|
| 637 |
-
"summary_jp": clean_text(raw.get("summary_jp")) or "้ขๆฅใๅฎไบใใพใใใ",
|
| 638 |
-
"total_questions": len(history),
|
| 639 |
-
"overall_score": clamp_int(raw.get("overall_score", 65), 0, 100),
|
| 640 |
-
"scores": {
|
| 641 |
-
"fluency": clamp_int(raw_scores.get("fluency", 6), 1, 10),
|
| 642 |
-
"grammar": clamp_int(raw_scores.get("grammar", 6), 1, 10),
|
| 643 |
-
"confidence": clamp_int(raw_scores.get("confidence", 6), 1, 10),
|
| 644 |
-
"relevance": clamp_int(raw_scores.get("relevance", 6), 1, 10),
|
| 645 |
-
"role_fit": clamp_int(raw_scores.get("role_fit", 6), 1, 10),
|
| 646 |
-
},
|
| 647 |
-
"pass_fail": "PASS" if str(raw.get("pass_fail", "PASS")).upper() == "PASS" else "FAIL",
|
| 648 |
-
"strengths": ensure_string_list(raw.get("strengths")),
|
| 649 |
-
"weaknesses": ensure_string_list(raw.get("weaknesses")),
|
| 650 |
-
"tips": ensure_string_list(raw.get("tips")),
|
| 651 |
-
"closing_message_jp": clean_text(raw.get("closing_message_jp")) or random_closing(role),
|
| 652 |
-
"answers": history,
|
| 653 |
-
}
|
| 654 |
-
|
| 655 |
-
|
| 656 |
-
def build_final_result(merged_memory: Dict[str, Any], role: Dict[str, Any], force_fail: bool = False, summary_jp: str = "") -> Dict[str, Any]:
|
| 657 |
-
history = merged_memory.get("answers_so_far", [])
|
| 658 |
-
scores = [int(item.get("answer_score", 0)) for item in history] or [0]
|
| 659 |
-
avg_10 = round(mean(scores), 1)
|
| 660 |
-
overall_score = clamp_int(avg_10 * 10, 0, 100)
|
| 661 |
-
if force_fail:
|
| 662 |
-
overall_score = min(overall_score, 39)
|
| 663 |
-
pass_fail = "PASS" if overall_score >= 60 and not force_fail else "FAIL"
|
| 664 |
-
|
| 665 |
-
strengths: List[str] = []
|
| 666 |
-
weaknesses: List[str] = []
|
| 667 |
-
tips: List[str] = []
|
| 668 |
-
|
| 669 |
-
if merged_memory.get("candidate_name"):
|
| 670 |
-
strengths.append("Basic self introduction was understood.")
|
| 671 |
-
else:
|
| 672 |
-
weaknesses.append("Name was not clearly understood.")
|
| 673 |
-
|
| 674 |
-
if merged_memory.get("experience_status") == "yes":
|
| 675 |
-
strengths.append("Candidate shared job-related experience.")
|
| 676 |
-
elif merged_memory.get("experience_status") == "no":
|
| 677 |
-
weaknesses.append("No direct experience was explained clearly.")
|
| 678 |
-
|
| 679 |
-
if overall_score >= 70:
|
| 680 |
-
strengths.append("Answers were mostly clear and relevant.")
|
| 681 |
-
else:
|
| 682 |
-
weaknesses.append("Several answers were short or unclear.")
|
| 683 |
-
|
| 684 |
-
tips.extend([
|
| 685 |
-
"Use one or two extra sentences in each answer.",
|
| 686 |
-
"Use polite endings like ใงใ and ใพใ.",
|
| 687 |
-
"Answer slowly and clearly.",
|
| 688 |
-
])
|
| 689 |
-
|
| 690 |
-
return {
|
| 691 |
-
"candidate_name": merged_memory.get("candidate_name"),
|
| 692 |
-
"country_name": merged_memory.get("country_name"),
|
| 693 |
-
"age": merged_memory.get("age"),
|
| 694 |
-
"job_role": merged_memory.get("job_role"),
|
| 695 |
-
"job_role_jp": merged_memory.get("job_role_jp"),
|
| 696 |
-
"summary_jp": summary_jp or "้ขๆฅใๅฎไบใใพใใใใใคใใใใพใงใใใ",
|
| 697 |
-
"total_questions": len(history),
|
| 698 |
-
"overall_score": overall_score,
|
| 699 |
-
"scores": {
|
| 700 |
-
"fluency": clamp_int(round(avg_10), 1, 10),
|
| 701 |
-
"grammar": clamp_int(round(avg_10 - 1), 1, 10),
|
| 702 |
-
"confidence": clamp_int(round(avg_10), 1, 10),
|
| 703 |
-
"relevance": clamp_int(round(avg_10 + 1), 1, 10),
|
| 704 |
-
"role_fit": clamp_int(round(avg_10), 1, 10),
|
| 705 |
-
},
|
| 706 |
-
"pass_fail": pass_fail,
|
| 707 |
-
"strengths": strengths,
|
| 708 |
-
"weaknesses": weaknesses,
|
| 709 |
-
"tips": tips,
|
| 710 |
-
"closing_message_jp": random_closing(role),
|
| 711 |
-
"answers": history,
|
| 712 |
-
}
|
| 713 |
-
|
| 714 |
-
|
| 715 |
-
def random_closing(role: Dict[str, Any]) -> str:
|
| 716 |
-
for q in role["questions"]:
|
| 717 |
-
if q["stage"] == "closing":
|
| 718 |
-
return q["jp"]
|
| 719 |
-
return f"ๆฌๆฅใฎ{role['japanese_name']}ใฎ้ขๆฅ็ทด็ฟใฏใใใพใงใงใใใๅๅ ใใใใจใใใใใพใใใ"
|
| 720 |
-
|
| 721 |
-
|
| 722 |
-
def choose_any_unused_question(role: Dict[str, Any], memory: Dict[str, Any]) -> Optional[Dict[str, Any]]:
|
| 723 |
-
asked_ids = set(memory.get("asked_question_ids", []))
|
| 724 |
-
for q in role["questions"]:
|
| 725 |
-
if q["id"] not in asked_ids and q["stage"] != "closing":
|
| 726 |
-
return q
|
| 727 |
-
return None
|
| 728 |
-
|
| 729 |
-
|
| 730 |
-
def get_question_by_id(role: Dict[str, Any], question_id: str) -> Optional[Dict[str, Any]]:
|
| 731 |
-
for q in role["questions"]:
|
| 732 |
-
if q["id"] == question_id:
|
| 733 |
-
return q
|
| 734 |
-
return None
|
| 735 |
-
|
| 736 |
-
|
| 737 |
def normalize_text(text: str) -> str:
|
| 738 |
return re.sub(r"\s+", " ", (text or "")).strip()
|
| 739 |
|
| 740 |
|
| 741 |
-
def clean_text(value: Any) -> str:
|
| 742 |
-
return normalize_text(str(value or ""))
|
| 743 |
-
|
| 744 |
-
|
| 745 |
-
def normalize_optional_text(value: Any) -> Optional[str]:
|
| 746 |
-
value = clean_text(value)
|
| 747 |
-
return value or None
|
| 748 |
-
|
| 749 |
-
|
| 750 |
-
def normalize_optional_int(value: Any) -> Optional[int]:
|
| 751 |
-
try:
|
| 752 |
-
if value in (None, ""):
|
| 753 |
-
return None
|
| 754 |
-
return int(value)
|
| 755 |
-
except Exception:
|
| 756 |
-
return None
|
| 757 |
-
|
| 758 |
-
|
| 759 |
-
def merge_unique(old_values: List[Any], new_values: List[Any]) -> List[Any]:
|
| 760 |
-
result = list(old_values or [])
|
| 761 |
-
for item in new_values or []:
|
| 762 |
-
if item not in result:
|
| 763 |
-
result.append(item)
|
| 764 |
-
return result
|
| 765 |
-
|
| 766 |
-
|
| 767 |
def safe_json_loads(value: str) -> Dict[str, Any]:
|
| 768 |
try:
|
| 769 |
parsed = json.loads(value or "{}")
|
| 770 |
return parsed if isinstance(parsed, dict) else {}
|
| 771 |
-
except json.JSONDecodeError:
|
| 772 |
-
return {}
|
| 773 |
-
|
| 774 |
-
|
| 775 |
-
def clamp_int(value: Any, low: int, high: int) -> int:
|
| 776 |
-
try:
|
| 777 |
-
return max(low, min(high, int(round(float(value)))))
|
| 778 |
except Exception:
|
| 779 |
-
return
|
| 780 |
-
|
| 781 |
-
|
| 782 |
-
def merge_memory(memory: Dict[str, Any], memory_update: Dict[str, Any]) -> Dict[str, Any]:
|
| 783 |
-
merged = dict(memory or {})
|
| 784 |
-
for key, value in (memory_update or {}).items():
|
| 785 |
-
if value not in (None, "", [], {}):
|
| 786 |
-
merged[key] = value
|
| 787 |
-
return merged
|
| 788 |
-
|
| 789 |
-
|
| 790 |
-
def maybe_extract_basic_profile(memory: Dict[str, Any], transcript: str, question_no: int) -> Dict[str, Any]:
|
| 791 |
-
update: Dict[str, Any] = {}
|
| 792 |
-
text = transcript.strip()
|
| 793 |
-
|
| 794 |
-
if not memory.get("candidate_name"):
|
| 795 |
-
name = extract_name(text)
|
| 796 |
-
if question_no <= 2 and name:
|
| 797 |
-
update["candidate_name"] = name
|
| 798 |
-
|
| 799 |
-
if not memory.get("country_name"):
|
| 800 |
-
country = extract_country(text)
|
| 801 |
-
if country:
|
| 802 |
-
update["country_name"] = country
|
| 803 |
-
|
| 804 |
-
if not memory.get("age"):
|
| 805 |
-
age = extract_age(text)
|
| 806 |
-
if age:
|
| 807 |
-
update["age"] = age
|
| 808 |
-
|
| 809 |
-
if not memory.get("reason_for_japan") and any(x in text for x in ["ๆฅๆฌ", "ๅใใใ", "่กใใใ", "ๅๅผท"]):
|
| 810 |
-
update["reason_for_japan"] = text[:80]
|
| 811 |
-
|
| 812 |
-
if not memory.get("occupation") and any(x in text for x in ["ไปไบ", "ๅใใฆ", "ๅญฆ็", "ๅๅผท"]):
|
| 813 |
-
update["occupation"] = text[:80]
|
| 814 |
-
|
| 815 |
-
if not memory.get("japanese_level") and any(x in text for x in ["ๆฅๆฌ่ช", "ๅๅผท", "ๅนด", "ใถๆ", "ๅฐใ"]):
|
| 816 |
-
update["japanese_level"] = text[:80]
|
| 817 |
-
|
| 818 |
-
update["experience_status"] = detect_experience_status(memory, text)
|
| 819 |
-
return update
|
| 820 |
-
|
| 821 |
-
|
| 822 |
-
def detect_experience_status(memory: Dict[str, Any], text: str) -> str:
|
| 823 |
-
current = str(memory.get("experience_status", "unknown"))
|
| 824 |
-
low = (text or "").strip()
|
| 825 |
-
yes_markers = ["ใใใพใ", "ใใพใใ", "ๅใใใใจใใใใพใ", "็ต้จใใใใพใ", "ใใฃใใใจใใใใพใ"]
|
| 826 |
-
no_markers = ["ใใใพใใ", "ใชใใงใ", "ใใใใจใใใใพใใ", "็ต้จใใใใพใใ", "ใชใ"]
|
| 827 |
-
if any(x in low for x in yes_markers):
|
| 828 |
-
return "yes"
|
| 829 |
-
if any(x in low for x in no_markers):
|
| 830 |
-
return "no"
|
| 831 |
-
return current if current in {"yes", "no"} else "unknown"
|
| 832 |
|
| 833 |
|
| 834 |
-
def extract_name(text: str) ->
|
| 835 |
value = text.replace("็งใฏ", "").replace("ใใใใฏ", "").replace("ใผใใฏ", "")
|
| 836 |
value = value.replace("ใงใ", "").replace("ใจ็ณใใพใ", "").replace("ใจใใใพใ", "").strip(" ใ")
|
| 837 |
-
|
| 838 |
-
return None
|
| 839 |
-
return value
|
| 840 |
-
|
| 841 |
-
|
| 842 |
-
def extract_country(text: str) -> Optional[str]:
|
| 843 |
-
known = ["ใใใผใซ", "ๆฅๆฌ", "ใคใณใ", "ใใณใฐใฉใใทใฅ", "ในใชใฉใณใซ", "ใใใใ ", "ไธญๅฝ", "ใใฃใณใใผ", "ใใฃใชใใณ", "ใคใณใใใทใข"]
|
| 844 |
-
for item in known:
|
| 845 |
-
if item in text:
|
| 846 |
-
return item
|
| 847 |
-
match = re.search(r"(.+?)ใใๆฅใพใใ", text)
|
| 848 |
-
if match:
|
| 849 |
-
return match.group(1).strip(" ใ")
|
| 850 |
-
return None
|
| 851 |
-
|
| 852 |
|
| 853 |
-
def extract_age(text: str) -> Optional[int]:
|
| 854 |
-
match = re.search(r"(\d{1,2})", text)
|
| 855 |
-
if match:
|
| 856 |
-
return int(match.group(1))
|
| 857 |
-
return None
|
| 858 |
|
| 859 |
-
|
| 860 |
-
|
| 861 |
-
hits = []
|
| 862 |
-
for kw in keywords:
|
| 863 |
-
if kw and kw in text:
|
| 864 |
-
hits.append(kw)
|
| 865 |
-
return hits
|
| 866 |
-
|
| 867 |
-
|
| 868 |
-
def heuristic_score(transcript: str, role: Dict[str, Any]) -> int:
|
| 869 |
-
text = transcript.strip()
|
| 870 |
if not text:
|
| 871 |
return 0
|
| 872 |
-
score =
|
| 873 |
if len(text) >= 6:
|
| 874 |
score += 1
|
| 875 |
if len(text) >= 12:
|
| 876 |
score += 1
|
| 877 |
if "ใงใ" in text or "ใพใ" in text:
|
| 878 |
score += 1
|
| 879 |
-
|
| 880 |
-
if hits >= 1:
|
| 881 |
-
score += 1
|
| 882 |
-
if hits >= 2:
|
| 883 |
-
score += 1
|
| 884 |
-
if len(text) >= 25:
|
| 885 |
score += 1
|
| 886 |
return min(score, 10)
|
| 887 |
|
|
@@ -894,14 +367,3 @@ def default_feedback(score: int) -> str:
|
|
| 894 |
if score >= 4:
|
| 895 |
return "ๆๅณใฏไผใใใพใใใๅฐใ็ญใใงใใๅฎๅ
จใชๆใง่ฉฑใใฆใฟใพใใใใ"
|
| 896 |
return "็ญใใใใใๅ
ๅฎนใๅใใใซใใใงใใใใๅฐใ่ฉณใใ่ฉฑใใฆใใ ใใใ"
|
| 897 |
-
|
| 898 |
-
|
| 899 |
-
def ensure_string_list(value: Any) -> List[str]:
|
| 900 |
-
if not isinstance(value, list):
|
| 901 |
-
return []
|
| 902 |
-
result = []
|
| 903 |
-
for item in value:
|
| 904 |
-
text = clean_text(item)
|
| 905 |
-
if text:
|
| 906 |
-
result.append(text)
|
| 907 |
-
return result
|
|
|
|
| 1 |
+
|
| 2 |
import json
|
| 3 |
import os
|
| 4 |
import re
|
|
|
|
|
|
|
| 5 |
from statistics import mean
|
| 6 |
+
from typing import Any, Dict, List
|
| 7 |
|
| 8 |
import requests
|
| 9 |
from fastapi import FastAPI, File, Form, UploadFile
|
| 10 |
from fastapi.middleware.cors import CORSMiddleware
|
| 11 |
from pydantic import BaseModel
|
| 12 |
|
| 13 |
+
app = FastAPI(title="Japanese AI Interview API", version="category-fix-1.0")
|
| 14 |
|
|
|
|
| 15 |
app.add_middleware(
|
| 16 |
CORSMiddleware,
|
| 17 |
allow_origins=["*"],
|
|
|
|
| 22 |
|
| 23 |
HF_TOKEN = os.getenv("HF_TOKEN", "").strip()
|
| 24 |
ASR_MODEL = os.getenv("ASR_MODEL", "openai/whisper-large-v3")
|
|
|
|
|
|
|
| 25 |
HF_INFERENCE_BASE = os.getenv("HF_INFERENCE_BASE", "https://router.huggingface.co/hf-inference/models")
|
|
|
|
|
|
|
|
|
|
| 26 |
|
| 27 |
+
ROLE_BANK: Dict[str, Dict[str, Any]] = {
|
| 28 |
+
"construction": {
|
| 29 |
+
"label": "Construction / ๅปบ่จญ",
|
| 30 |
+
"intro": "ใใใซใกใฏใๅปบ่จญใฎไปไบใฎ้ขๆฅ็ทด็ฟใๅงใใพใใใใใใใ้กใใใพใใ",
|
| 31 |
+
"questions": [
|
| 32 |
+
"ใๅๅใๆใใฆใใ ใใใ",
|
| 33 |
+
"ใฉใใฎๅฝใใๆฅใพใใใใ",
|
| 34 |
+
"ๆฅๆฌใธ่กใใใ็็ฑใฏไฝใงใใใ",
|
| 35 |
+
"ๅปบ่จญใฎไปไบใใใใใจใใใใพใใใ",
|
| 36 |
+
"ๅฑใชใๅ ดๆใงๅใใจใใไฝใซๆฐใใคใใพใใใ",
|
| 37 |
+
"ใใผใ ใงๅใใใจใฏใงใใพใใใ",
|
| 38 |
+
"ไฝๅใซ่ชไฟกใฏใใใพใใใ",
|
| 39 |
+
"ๆๅพใซใๅปบ่จญใฎไปไบใงใใใฐใใใใใจใ่ฉฑใใฆใใ ใใใ"
|
| 40 |
+
]
|
| 41 |
+
},
|
| 42 |
+
"restaurant_konbini": {
|
| 43 |
+
"label": "Restaurant / Konbini / ๅค้ฃใปใณใณใใ",
|
| 44 |
+
"intro": "ใใใซใกใฏใๅค้ฃใปใณใณใใใฎไปไบใฎ้ขๆฅ็ทด็ฟใๅงใใพใใใใใใใ้กใใใพใใ",
|
| 45 |
+
"questions": [
|
| 46 |
+
"ใๅๅใๆใใฆใใ ใใใ",
|
| 47 |
+
"ใฉใใฎๅฝใใๆฅใพใใใใ",
|
| 48 |
+
"ๆฅๆฌใธ่กใใใ็็ฑใฏไฝใงใใใ",
|
| 49 |
+
"ๆฅๅฎขใฎไปไบใใใใใจใใใใพใใใ",
|
| 50 |
+
"ใๅฎขๆงใซใฏใฉใฎใใใซ่ฉฑใใพใใใ",
|
| 51 |
+
"ๅฟใใๆ้ใงใ่ฝใก็ใใฆๅใใพใใใ",
|
| 52 |
+
"็ฌ้กใงๆฅๅฎขใงใใพใใใ",
|
| 53 |
+
"ๆๅพใซใใใฎไปไบใซๅใใฆใใ่ชๅใฎ้ทๆใ่ฉฑใใฆใใ ใใใ"
|
| 54 |
+
]
|
| 55 |
+
},
|
| 56 |
+
"nursing_care": {
|
| 57 |
+
"label": "Nursing Care / ไป่ญท",
|
| 58 |
+
"intro": "ใใใซใกใฏใไป่ญทใฎไปไบใฎ้ขๆฅ็ทด็ฟใๅงใใพใใใใใใใ้กใใใพใใ",
|
| 59 |
+
"questions": [
|
| 60 |
+
"ใๅๅใๆใใฆใใ ใใใ",
|
| 61 |
+
"ใฉใใฎๅฝใใๆฅใพใใใใ",
|
| 62 |
+
"ๆฅๆฌใธ่กใใใ็็ฑใฏไฝใงใใใ",
|
| 63 |
+
"ไป่ญทใฎไปไบใใใใใจใใใใพใใใ",
|
| 64 |
+
"ใๅนดๅฏใใๅฉ็จ่
ใใใซใใใใ่ฉฑใใพใใใ",
|
| 65 |
+
"ไป่ญทใฎไปไบใงๆธ
ๆฝใใฏๅคงๅใงใใใใชใใงใใใ",
|
| 66 |
+
"ๅฉ็จ่
ใใใๅฐใฃใฆใใใใใฉใใใพใใใ",
|
| 67 |
+
"ๆๅพใซใไป่ญทใฎไปไบใงๅคงๅใ ใจๆใใใจใ่ฉฑใใฆใใ ใใใ"
|
| 68 |
+
]
|
| 69 |
+
},
|
| 70 |
+
"hotel_accommodation": {
|
| 71 |
+
"label": "Hotel / Accommodation / ๅฎฟๆณ",
|
| 72 |
+
"intro": "ใใใซใกใฏใๅฎฟๆณใฎไปไบใฎ้ขๆฅ็ทด็ฟใๅงใใพใใใใใใใ้กใใใพใใ",
|
| 73 |
+
"questions": [
|
| 74 |
+
"ใๅๅใๆใใฆใใ ใใใ",
|
| 75 |
+
"ใฉใใฎๅฝใใๆฅใพใใใใ",
|
| 76 |
+
"ๆฅๆฌใธ่กใใใ็็ฑใฏไฝใงใใใ",
|
| 77 |
+
"ใใใซใฎไปไบใใใใใจใใใใพใใใ",
|
| 78 |
+
"ใๅฎขๆงใซใฆใใญใใซ่ฉฑใใพใใใ",
|
| 79 |
+
"ๆ้คใใใใใกใคใฏใฏใงใใพใใใ",
|
| 80 |
+
"ๅฟใใๆ้ใงใ่ฝใก็ใใฆๅใใพใใใ",
|
| 81 |
+
"ๆๅพใซใใใใซใฎไปไบใง่ชๅใฎ้ทๆใ่ฉฑใใฆใใ ใใใ"
|
| 82 |
+
]
|
| 83 |
+
},
|
| 84 |
+
"agriculture": {
|
| 85 |
+
"label": "Agriculture / ่พฒๆฅญ",
|
| 86 |
+
"intro": "ใใใซใกใฏใ่พฒๆฅญใฎไปไบใฎ้ขๆฅ็ทด็ฟใๅงใใพใใใใใใใ้กใใใพใใ",
|
| 87 |
+
"questions": [
|
| 88 |
+
"ใๅๅใๆใใฆใใ ใใใ",
|
| 89 |
+
"ใฉใใฎๅฝใใๆฅใพใใใใ",
|
| 90 |
+
"ๆฅๆฌใธ่กใใใ็็ฑใฏไฝใงใใใ",
|
| 91 |
+
"่พฒๆฅญใฎไปไบใใใใใจใใใใพใใใ",
|
| 92 |
+
"ๅคใงๅใใใจใฏๅคงไธๅคซใงใใใ",
|
| 93 |
+
"ๆๆฉใไปไบใงใๆ้ใๅฎใใพใใใ",
|
| 94 |
+
"ไฝๅใซ่ชไฟกใฏใใใพใใใ",
|
| 95 |
+
"ๆๅพใซใ่พฒๆฅญใฎไปไบใงใใใฐใใใใใจใ่ฉฑใใฆใใ ใใใ"
|
| 96 |
+
]
|
| 97 |
+
},
|
| 98 |
+
"manufacturing": {
|
| 99 |
+
"label": "Manufacturing / ่ฃฝ้ ๆฅญ",
|
| 100 |
+
"intro": "ใใใซใกใฏใ่ฃฝ้ ๆฅญใฎไปไบใฎ้ขๆฅ็ทด็ฟใๅงใใพใใใใใใใ้กใใใพใใ",
|
| 101 |
+
"questions": [
|
| 102 |
+
"ใๅๅใๆใใฆใใ ใใใ",
|
| 103 |
+
"ใฉใใฎๅฝใใๆฅใพใใใใ",
|
| 104 |
+
"ๆฅๆฌใธ่กใใใ็็ฑใฏไฝใงใใใ",
|
| 105 |
+
"ๅทฅๅ ดใฎไปไบใใใใใจใใใใพใใใ",
|
| 106 |
+
"ๅฎๅ
จใซใผใซใๅฎใใใจใฏใงใใพใใใ",
|
| 107 |
+
"ๅใไฝๆฅญใๆญฃ็ขบใซ็ถใใใใพใใใ",
|
| 108 |
+
"ๆ้ใๅฎใใใจใฏใงใใพใใใ",
|
| 109 |
+
"ๆๅพใซใ่ฃฝ้ ๆฅญใฎไปไบใง่ชๅใฎ้ทๆใ่ฉฑใใฆใใ ใใใ"
|
| 110 |
+
]
|
| 111 |
+
},
|
| 112 |
+
}
|
| 113 |
+
|
| 114 |
+
REPEAT_PROMPT = "ๅฃฐใๅฐใใใงใใใใๅฐใๅคงใใๅฃฐใงใใใไธๅบฆใ้กใใใพใใ"
|
| 115 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 116 |
|
| 117 |
class StartRequest(BaseModel):
|
| 118 |
session_uuid: str
|
| 119 |
job_role: str = "construction"
|
|
|
|
| 120 |
|
| 121 |
|
| 122 |
@app.get("/")
|
| 123 |
def root() -> Dict[str, Any]:
|
| 124 |
+
return {"ok": True, "service": "jp-interview", "version": "category-fix-1.0"}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 125 |
|
| 126 |
|
| 127 |
@app.get("/health")
|
| 128 |
def health() -> Dict[str, Any]:
|
| 129 |
return {
|
| 130 |
"ok": True,
|
| 131 |
+
"service": "jp-interview",
|
| 132 |
+
"version": "category-fix-1.0",
|
| 133 |
"hf_token_set": bool(HF_TOKEN),
|
| 134 |
"asr_model": ASR_MODEL,
|
| 135 |
+
"roles": list(ROLE_BANK.keys()),
|
|
|
|
|
|
|
|
|
|
| 136 |
}
|
| 137 |
|
| 138 |
|
| 139 |
@app.get("/roles")
|
| 140 |
def roles() -> Dict[str, Any]:
|
| 141 |
+
return {
|
| 142 |
+
"ok": True,
|
| 143 |
+
"roles": [{"key": key, "label": cfg["label"]} for key, cfg in ROLE_BANK.items()]
|
| 144 |
+
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 145 |
|
| 146 |
|
| 147 |
@app.post("/start")
|
| 148 |
def start_interview(payload: StartRequest) -> Dict[str, Any]:
|
| 149 |
role_key = payload.job_role if payload.job_role in ROLE_BANK else "construction"
|
| 150 |
+
cfg = ROLE_BANK[role_key]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 151 |
memory = {
|
|
|
|
| 152 |
"job_role": role_key,
|
| 153 |
+
"job_role_label": cfg["label"],
|
|
|
|
|
|
|
| 154 |
"candidate_name": None,
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 155 |
"answers_so_far": [],
|
|
|
|
|
|
|
| 156 |
"low_score_count": 0,
|
| 157 |
"no_sound_count": 0,
|
| 158 |
+
"question_count_mode": "auto",
|
|
|
|
|
|
|
| 159 |
}
|
| 160 |
+
opening = f"{cfg['intro']} {cfg['questions'][0]}"
|
| 161 |
return {
|
| 162 |
"ok": True,
|
| 163 |
"session_uuid": payload.session_uuid,
|
| 164 |
"job_role": role_key,
|
| 165 |
+
"job_role_label": cfg["label"],
|
| 166 |
"question_no": 1,
|
| 167 |
+
"question_jp": cfg["questions"][0],
|
| 168 |
+
"speech_text_jp": opening,
|
|
|
|
| 169 |
"memory": memory,
|
| 170 |
"is_finished": False,
|
| 171 |
"speak_now": True,
|
|
|
|
| 176 |
async def answer_interview(
|
| 177 |
session_uuid: str = Form(...),
|
| 178 |
question_no: int = Form(...),
|
|
|
|
|
|
|
| 179 |
question_jp: str = Form(...),
|
| 180 |
memory_json: str = Form("{}"),
|
| 181 |
audio: UploadFile = File(...),
|
| 182 |
) -> Dict[str, Any]:
|
| 183 |
memory = safe_json_loads(memory_json)
|
| 184 |
+
role_key = memory.get("job_role") if memory.get("job_role") in ROLE_BANK else "construction"
|
| 185 |
+
cfg = ROLE_BANK[role_key]
|
|
|
|
| 186 |
|
| 187 |
audio_bytes = await audio.read()
|
| 188 |
transcript = ""
|
| 189 |
+
if audio_bytes and HF_TOKEN:
|
|
|
|
| 190 |
try:
|
| 191 |
transcript = transcribe_audio_with_hf(audio_bytes, audio.filename or "audio.webm")
|
| 192 |
+
except Exception:
|
| 193 |
+
transcript = ""
|
| 194 |
|
| 195 |
if not transcript.strip():
|
| 196 |
memory["no_sound_count"] = int(memory.get("no_sound_count", 0)) + 1
|
| 197 |
+
speech_text = REPEAT_PROMPT
|
| 198 |
+
if memory.get("candidate_name"):
|
| 199 |
+
speech_text = f"{memory['candidate_name']}ใใใ{REPEAT_PROMPT}"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 200 |
return {
|
| 201 |
"ok": True,
|
| 202 |
"is_finished": False,
|
| 203 |
"needs_repeat": True,
|
| 204 |
"session_uuid": session_uuid,
|
| 205 |
"question_no": question_no,
|
|
|
|
|
|
|
| 206 |
"question_jp": question_jp,
|
| 207 |
+
"speech_text_jp": speech_text,
|
| 208 |
"transcript_jp": "",
|
| 209 |
"answer_score": 0,
|
| 210 |
+
"feedback_jp": REPEAT_PROMPT,
|
| 211 |
"memory": memory,
|
| 212 |
"next_question_no": question_no,
|
|
|
|
| 213 |
"next_question_jp": question_jp,
|
| 214 |
"speak_now": True,
|
|
|
|
|
|
|
| 215 |
}
|
| 216 |
|
| 217 |
+
if question_no == 1 and not memory.get("candidate_name"):
|
| 218 |
+
name = extract_name(transcript)
|
| 219 |
+
if name:
|
| 220 |
+
memory["candidate_name"] = name
|
| 221 |
|
| 222 |
+
score = heuristic_score(transcript)
|
| 223 |
+
if score <= 3:
|
| 224 |
+
memory["low_score_count"] = int(memory.get("low_score_count", 0)) + 1
|
| 225 |
+
else:
|
| 226 |
+
memory["low_score_count"] = 0
|
| 227 |
+
|
| 228 |
+
history = list(memory.get("answers_so_far", []))
|
| 229 |
+
history.append({
|
| 230 |
"question_no": question_no,
|
|
|
|
| 231 |
"question_jp": question_jp,
|
| 232 |
"answer_text_jp": transcript,
|
| 233 |
+
"answer_score": score,
|
| 234 |
+
})
|
| 235 |
+
memory["answers_so_far"] = history
|
| 236 |
+
|
| 237 |
+
# automatic interview length
|
| 238 |
+
min_questions = 3
|
| 239 |
+
max_questions = len(cfg["questions"])
|
| 240 |
+
avg_score = mean([x.get("answer_score", 0) for x in history])
|
| 241 |
+
|
| 242 |
+
finish_now = False
|
| 243 |
+
if question_no >= min_questions and memory["low_score_count"] >= 2:
|
| 244 |
+
finish_now = True
|
| 245 |
+
elif question_no >= min_questions and avg_score < 3.5:
|
| 246 |
+
finish_now = True
|
| 247 |
+
elif question_no >= max_questions:
|
| 248 |
+
finish_now = True
|
| 249 |
+
elif question_no >= 5 and avg_score < 5:
|
| 250 |
+
finish_now = True
|
| 251 |
+
|
| 252 |
+
if finish_now:
|
| 253 |
+
closing = f"ๆฌๆฅใฎ{cfg['label'].split('/')[1].strip() if '/' in cfg['label'] else cfg['label']}ใฎ้ขๆฅ็ทด็ฟใฏใใใพใงใงใใใๅๅ ใใใใจใใใใใพใใใ"
|
| 254 |
+
result = {
|
| 255 |
+
"candidate_name": memory.get("candidate_name"),
|
| 256 |
+
"job_role": role_key,
|
| 257 |
+
"job_role_label": cfg["label"],
|
| 258 |
+
"total_questions": len(history),
|
| 259 |
+
"overall_score": round(avg_score * 10),
|
| 260 |
+
"pass_fail": "PASS" if avg_score >= 6 else "FAIL",
|
| 261 |
+
"closing_message_jp": closing,
|
| 262 |
+
"answers": history,
|
| 263 |
+
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 264 |
return {
|
| 265 |
"ok": True,
|
| 266 |
"is_finished": True,
|
| 267 |
"session_uuid": session_uuid,
|
| 268 |
"question_no": question_no,
|
|
|
|
| 269 |
"transcript_jp": transcript,
|
| 270 |
+
"answer_score": score,
|
| 271 |
+
"feedback_jp": default_feedback(score),
|
| 272 |
+
"speech_text_jp": closing,
|
| 273 |
+
"memory": memory,
|
| 274 |
"result": result,
|
| 275 |
}
|
| 276 |
|
| 277 |
next_question_no = question_no + 1
|
| 278 |
+
next_question = cfg["questions"][next_question_no - 1]
|
| 279 |
+
speech_text = next_question
|
| 280 |
+
if question_no == 1 and memory.get("candidate_name"):
|
| 281 |
+
speech_text = f"{memory['candidate_name']}ใใใใใใใจใใใใใพใใ้ขๆฅใฎๆบๅใฏใงใใฆใใพใใใใงใฏใๆฌกใฎ่ณชๅใงใใ{next_question}"
|
|
|
|
|
|
|
|
|
|
| 282 |
|
| 283 |
return {
|
| 284 |
"ok": True,
|
| 285 |
"is_finished": False,
|
| 286 |
"session_uuid": session_uuid,
|
| 287 |
"question_no": question_no,
|
|
|
|
| 288 |
"transcript_jp": transcript,
|
| 289 |
+
"answer_score": score,
|
| 290 |
+
"feedback_jp": default_feedback(score),
|
| 291 |
+
"speech_text_jp": speech_text,
|
| 292 |
+
"memory": memory,
|
| 293 |
"next_question_no": next_question_no,
|
| 294 |
+
"next_question_jp": next_question,
|
|
|
|
| 295 |
"speak_now": True,
|
| 296 |
}
|
| 297 |
|
| 298 |
|
| 299 |
def transcribe_audio_with_hf(audio_bytes: bytes, filename: str) -> str:
|
|
|
|
|
|
|
| 300 |
url = f"{HF_INFERENCE_BASE}/{ASR_MODEL}"
|
| 301 |
headers = {
|
| 302 |
"Authorization": f"Bearer {HF_TOKEN}",
|
| 303 |
"Content-Type": guess_mime_type(filename),
|
| 304 |
}
|
| 305 |
+
response = requests.post(url, headers=headers, data=audio_bytes, timeout=180)
|
| 306 |
response.raise_for_status()
|
| 307 |
data = response.json()
|
| 308 |
if isinstance(data, dict):
|
| 309 |
+
return normalize_text(data.get("text") or data.get("generated_text") or "")
|
| 310 |
+
if isinstance(data, list) and data and isinstance(data[0], dict):
|
| 311 |
+
return normalize_text(data[0].get("text", ""))
|
|
|
|
|
|
|
|
|
|
| 312 |
return ""
|
| 313 |
|
| 314 |
|
|
|
|
| 325 |
return "audio/webm"
|
| 326 |
|
| 327 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 328 |
def normalize_text(text: str) -> str:
|
| 329 |
return re.sub(r"\s+", " ", (text or "")).strip()
|
| 330 |
|
| 331 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 332 |
def safe_json_loads(value: str) -> Dict[str, Any]:
|
| 333 |
try:
|
| 334 |
parsed = json.loads(value or "{}")
|
| 335 |
return parsed if isinstance(parsed, dict) else {}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 336 |
except Exception:
|
| 337 |
+
return {}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 338 |
|
| 339 |
|
| 340 |
+
def extract_name(text: str) -> str:
|
| 341 |
value = text.replace("็งใฏ", "").replace("ใใใใฏ", "").replace("ใผใใฏ", "")
|
| 342 |
value = value.replace("ใงใ", "").replace("ใจ็ณใใพใ", "").replace("ใจใใใพใ", "").strip(" ใ")
|
| 343 |
+
return value[:30] if value else ""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 344 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 345 |
|
| 346 |
+
def heuristic_score(text: str) -> int:
|
| 347 |
+
text = (text or "").strip()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 348 |
if not text:
|
| 349 |
return 0
|
| 350 |
+
score = 4
|
| 351 |
if len(text) >= 6:
|
| 352 |
score += 1
|
| 353 |
if len(text) >= 12:
|
| 354 |
score += 1
|
| 355 |
if "ใงใ" in text or "ใพใ" in text:
|
| 356 |
score += 1
|
| 357 |
+
if len(text) >= 20:
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 358 |
score += 1
|
| 359 |
return min(score, 10)
|
| 360 |
|
|
|
|
| 367 |
if score >= 4:
|
| 368 |
return "ๆๅณใฏไผใใใพใใใๅฐใ็ญใใงใใๅฎๅ
จใชๆใง่ฉฑใใฆใฟใพใใใใ"
|
| 369 |
return "็ญใใใใใๅ
ๅฎนใๅใใใซใใใงใใใใๅฐใ่ฉณใใ่ฉฑใใฆใใ ใใใ"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|