cesnimda/jobtrackingapp
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: add application package generation and grouped readiness workflows

Browse Source
This commit is contained in:
cesnimda
2026-03-22 18:28:02 +01:00
parent f1c7c38a19
commit 9188039e9d
14 changed files with 1014 additions and 373 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tools/summarizer/app.py
+69 -71
View File
@@ -11,25 +11,24 @@ app = FastAPI(title="Local Summarizer")
MODEL_NAME = "sshleifer/distilbart-cnn-12-6"
MAX_INPUT_CHARS = 20000
# The local summarizer is intentionally simple, but we still validate request sizes
# so accidental giant pastes do not cause avoidable latency or memory spikes.
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
model = AutoModelForSeq2SeqLM.from_pretrained(MODEL_NAME)
model.eval()
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)
cache = TTLCache(maxsize=1024, ttl=60 * 60) # 1 hour cache
cache = TTLCache(maxsize=1024, ttl=60 * 60)
class SummarizeRequest(BaseModel):
text: str = Field(min_length=1, max_length=MAX_INPUT_CHARS)
max_length: int = Field(default=160, ge=24, le=256)
min_length: int = Field(default=45, ge=8, le=180)
top_skills: int = Field(default=8, ge=3, le=12)
def _key(text: str, max_length: int, min_length: int) -> str:
def _key(text: str, max_length: int, min_length: int, top_skills: int) -> str:
h = hashlib.sha256(text.encode("utf-8")).hexdigest()
return f"{h}:{max_length}:{min_length}"
return f"{h}:{max_length}:{min_length}:{top_skills}"
@app.get("/health")
@@ -38,52 +37,39 @@ async def health():
_TECH = [
"python",
"c#",
"dotnet",
".net",
"java",
"javascript",
"typescript",
"react",
"node",
"sql",
"postgres",
"postgresql",
"mysql",
"sqlite",
"mongodb",
"redis",
"aws",
"azure",
"gcp",
"docker",
"kubernetes",
"terraform",
"linux",
"git",
"ci/cd",
"graphql",
"rest",
"python", "c#", "dotnet", ".net", "java", "javascript", "typescript", "react", "node", "sql",
"postgres", "postgresql", "mysql", "sqlite", "mongodb", "redis", "aws", "azure", "gcp",
"docker", "kubernetes", "terraform", "linux", "git", "ci/cd", "graphql", "rest",
]
_SOFT = [
"communication",
"collaboration",
"teamwork",
"problem solving",
"leadership",
"mentoring",
"ownership",
"initiative",
"adaptability",
"stakeholder management",
"detail oriented",
"communication", "collaboration", "teamwork", "problem solving", "leadership", "mentoring",
"ownership", "initiative", "adaptability", "stakeholder management", "detail oriented",
]
_TECH_PRIORITY = [
"python", "c#", ".net", "dotnet", "typescript", "javascript", "react", "node",
"sql", "postgresql", "postgres", "mysql", "sqlite", "docker", "kubernetes",
"aws", "azure", "gcp", "terraform", "graphql", "rest", "git",
]
def _rank_tech_skills(skills):
ordered = []
seen = set()
for preferred in _TECH_PRIORITY:
for skill in skills:
if skill == preferred and skill not in seen:
ordered.append(skill)
seen.add(skill)
for skill in skills:
if skill not in seen:
ordered.append(skill)
seen.add(skill)
return ordered
def _strip_html(text: str) -> str:
# Good enough for job descriptions pasted from the web.
text = re.sub(r"<\s*br\s*/?>", "\n", text, flags=re.IGNORECASE)
text = re.sub(r"</p\s*>", "\n", text, flags=re.IGNORECASE)
text = re.sub(r"<[^>]+>", " ", text)
@@ -105,6 +91,21 @@ def _extract_bullets(lines, max_items=8):
return out
def _top_keywords(text: str, limit=6):
words = re.findall(r"[a-zA-Z][a-zA-Z+#./-]{2,}", text.lower())
stop = {
"with", "from", "that", "this", "will", "have", "your", "their", "about", "role", "team", "work",
"experience", "skills", "requirements", "responsibilities", "company", "using", "ability", "years",
}
counts = {}
for word in words:
if word in stop or word in _TECH or word in _SOFT:
continue
counts[word] = counts.get(word, 0) + 1
ordered = sorted(counts.items(), key=lambda item: (-item[1], item[0]))
return [word for word, _ in ordered[:limit]]
def _role_focused_excerpt(text: str) -> dict:
cleaned = _strip_html(text)
lines = [ln.strip() for ln in cleaned.splitlines()]
@@ -117,10 +118,10 @@ def _role_focused_excerpt(text: str) -> dict:
def match_heading(s: str):
sl = s.lower().strip(":-\x7f ")
for k, words in headings.items():
for w in words:
if sl == w or sl.startswith(w + " "):
return k
for key, words in headings.items():
for word in words:
if sl == word or sl.startswith(word + " "):
return key
return None
section = None
@@ -131,11 +132,10 @@ def _role_focused_excerpt(text: str) -> dict:
for ln in lines:
if not ln:
continue
h = match_heading(ln)
if h:
section = h
heading = match_heading(ln)
if heading:
section = heading
continue
if section == "responsibilities":
resp_lines.append(ln)
elif section == "requirements":
@@ -157,7 +157,6 @@ def _role_focused_excerpt(text: str) -> dict:
if s in low:
soft_found.append(s)
# Fallback: pick bullet-like lines anywhere if sections are missing.
if not responsibilities and not requirements:
any_bullets = _extract_bullets(lines, max_items=10)
responsibilities = any_bullets[:6]
@@ -170,8 +169,6 @@ def _role_focused_excerpt(text: str) -> dict:
focused_parts.append("Requirements:\n- " + "\n- ".join(requirements))
if nice:
focused_parts.append("Nice to have:\n- " + "\n- ".join(nice))
# Always include a small slice of the original for context.
focused_parts.append("Context:\n" + cleaned[:1500])
return {
@@ -182,6 +179,7 @@ def _role_focused_excerpt(text: str) -> dict:
"nice": nice,
"tech": tech_found,
"soft": soft_found,
"keywords": _top_keywords(cleaned),
}
@@ -206,20 +204,18 @@ async def summarize(req: SummarizeRequest):
if req.min_length >= req.max_length:
raise HTTPException(status_code=400, detail="min_length must be smaller than max_length.")
key = _key(req.text, req.max_length, req.min_length)
key = _key(req.text, req.max_length, req.min_length, req.top_skills)
if key in cache:
return {"summary": cache[key], "cached": True}
info = _role_focused_excerpt(req.text)
# Summarize the role-focused excerpt instead of the whole job post.
summary = _model_summarize(info["focused_input"], req.max_length, req.min_length)
lines = ["Role summary:", summary]
if info["requirements"]:
lines.append("")
lines.append("What they need from you:")
lines.append("What the company wants most:")
for x in info["requirements"][:7]:
lines.append(f"- {x}")
@@ -241,7 +237,7 @@ async def summarize(req: SummarizeRequest):
if t not in uniq:
uniq.append(t)
lines.append("")
lines.append("Top hard skills: " + ", ".join(uniq[:10]))
lines.append("Top hard skills: " + ", ".join(uniq[: req.top_skills]))
if info["soft"]:
uniq_soft = []
@@ -251,18 +247,20 @@ async def summarize(req: SummarizeRequest):
lines.append("")
lines.append("Relevant soft skills: " + ", ".join(uniq_soft[:8]))
out = "\n".join(lines).strip()
cache[key] = out
return {"summary": out, "cached": False}
skills: " + ", ".join(uniq[:14]))
if info["soft"]:
uniq_soft = []
for s in info["soft"]:
if s not in uniq_soft:
uniq_soft.append(s)
if info["keywords"]:
lines.append("")
lines.append("Relevant soft skills: " + ", ".join(uniq_soft[:8]))
lines.append("Key themes: " + ", ".join(info["keywords"][:6]))
lines.append("")
lines.append("Interview focus:")
if info["requirements"]:
for x in info["requirements"][:3]:
lines.append(f"- Prepare examples that demonstrate: {x}")
elif info["tech"]:
for x in _rank_tech_skills(info["tech"])[:3]:
lines.append(f"- Be ready to explain your hands-on experience with {x}")
else:
lines.append("- Prepare examples showing relevant impact, collaboration, and delivery.")
out = "\n".join(lines).strip()
cache[key] = out