offerprinter 0.2.0__py3-none-any.whl

offerprinter/models/schemas.py

@@ -0,0 +1,202 @@
+"""Data models for OfferPrinter.
+
+These are deliberately thin, typed containers (pydantic) that carry data between
+the layers — config → services → controller → output. Keeping them in one place
+makes the flow easy to read and extend.
+"""
+
+from __future__ import annotations
+
+from enum import StrEnum
+
+from pydantic import BaseModel, Field
+
+
+class Provider(StrEnum):
+    """Supported LLM providers."""
+
+    ANTHROPIC = "anthropic"
+    OPENAI = "openai"
+    GEMINI = "gemini"
+    KIMI = "kimi"
+    OLLAMA = "ollama"
+
+
+class Locale(StrEnum):
+    """Output spelling/idiom."""
+
+    UK = "UK"
+    US = "US"
+
+
+class LLMConfig(BaseModel):
+    """Everything the LLM layer needs to make a call."""
+
+    provider: Provider = Provider.ANTHROPIC
+    model: str = ""  # "" means "use the provider default"
+    api_key: str = ""
+    base_url: str = ""
+    temperature: float = 0.2
+    max_tokens: int = 4096
+    timeout: float = 120.0
+    max_retries: int = 3  # retries on 429 / 5xx / network blips
+    retry_backoff: float = 1.5  # seconds; doubles each attempt
+
+
+class OutputConfig(BaseModel):
+    """Where and how results are written."""
+
+    locale: Locale = Locale.UK
+    dir: str = "./output"
+    formats: list[str] = Field(default_factory=lambda: ["md", "docx"])
+    #: Append each run to the local history in ~/.offerprinter/applications.json.
+    track: bool = True
+
+
+class GenerationConfig(BaseModel):
+    """Which artifacts to produce, and how fast to produce them."""
+
+    tailored_cv: bool = True
+    cover_letter: bool = True
+    fit_memo: bool = True
+    ats_report: bool = True
+    interview_prep: bool = True
+
+    #: Score the application 0-100 and band it. One extra cheap call.
+    fit_score: bool = True
+    #: Generate all artifacts concurrently instead of one after another.
+    parallel: bool = True
+    #: Upper bound on concurrent LLM calls.
+    max_workers: int = 5
+
+    def enabled(self) -> list[str]:
+        """Return the keys of the artifacts that are switched on, in order."""
+        return [k for k in ARTIFACT_ORDER if getattr(self, k)]
+
+
+#: Canonical artifact order — used for output ordering regardless of which
+#: artifact finishes first when generating in parallel.
+ARTIFACT_ORDER: list[str] = [
+    "tailored_cv",
+    "cover_letter",
+    "fit_memo",
+    "ats_report",
+    "interview_prep",
+]
+
+
+class ResumeInput(BaseModel):
+    """The user's base CV, already extracted to plain text."""
+
+    text: str
+    source: str = "pasted"  # a filename or "pasted"
+
+
+class JobDescription(BaseModel):
+    """The target job, already extracted to plain text."""
+
+    text: str
+    source: str = "pasted"  # a URL or "pasted"
+    company: str = ""  # best-effort; filled in by the pipeline
+    role: str = ""  # best-effort; filled in by the pipeline
+
+
+class Artifact(BaseModel):
+    """One generated document."""
+
+    key: str  # e.g. "tailored_cv"
+    title: str  # human title, e.g. "Tailored CV"
+    filename: str  # base filename without extension, e.g. "tailored-cv"
+    content: str  # markdown body
+
+
+class Usage(BaseModel):
+    """Token counts and estimated spend for one run."""
+
+    input_tokens: int = 0
+    output_tokens: int = 0
+    calls: int = 0
+    cost_usd: float = 0.0
+
+    @property
+    def total_tokens(self) -> int:
+        return self.input_tokens + self.output_tokens
+
+    def add(self, other: Usage) -> None:
+        self.input_tokens += other.input_tokens
+        self.output_tokens += other.output_tokens
+        self.calls += other.calls
+        self.cost_usd += other.cost_usd
+
+
+#: Score bands, highest threshold first. (min_score, label, one-line verdict)
+FIT_BANDS: list[tuple[int, str, str]] = [
+    (85, "Exceptional", "Apply today. You are what they wrote the advert for."),
+    (70, "Strong", "A genuinely competitive application. Send it."),
+    (55, "Credible", "Worth applying — lead hard with your strongest match."),
+    (40, "Stretch", "A reach. Apply if you want it, and address the gaps head-on."),
+    (0, "Long shot", "Big gaps. Consider a closer role, or close a gap first."),
+]
+
+
+class FitScore(BaseModel):
+    """How well the candidate genuinely matches the role."""
+
+    score: int = 0  # 0-100
+    band: str = ""
+    verdict: str = ""
+    strengths: list[str] = Field(default_factory=list)
+    gaps: list[str] = Field(default_factory=list)
+
+    @classmethod
+    def band_for(cls, score: int) -> tuple[str, str]:
+        """Return (band, verdict) for a 0-100 score."""
+        for threshold, label, verdict in FIT_BANDS:
+            if score >= threshold:
+                return label, verdict
+        return FIT_BANDS[-1][1], FIT_BANDS[-1][2]
+
+    def as_markdown(self) -> str:
+        filled = round(self.score / 10)
+        bar = "█" * filled + "░" * (10 - filled)
+        lines = [
+            "# Fit Score",
+            "",
+            f"## {self.score}/100 — {self.band}",
+            "",
+            f"`{bar}`",
+            "",
+            f"*{self.verdict}*",
+            "",
+        ]
+        if self.strengths:
+            lines += ["## Where you genuinely match", ""]
+            lines += [f"- {s}" for s in self.strengths] + [""]
+        if self.gaps:
+            lines += ["## Real gaps (not filled in for you)", ""]
+            lines += [f"- {g}" for g in self.gaps] + [""]
+        lines += [
+            "> Scored from evidence actually present in your CV. Gaps are stated,",
+            "> never papered over. — OfferPrinter",
+            "",
+        ]
+        return "\n".join(lines)
+
+
+class ApplicationPackage(BaseModel):
+    """The full result of one run."""
+
+    company: str
+    role: str
+    slug: str  # <company>-<role> folder name
+    artifacts: list[Artifact] = Field(default_factory=list)
+    fit: FitScore | None = None
+    usage: Usage = Field(default_factory=Usage)
+
+    def get(self, key: str) -> Artifact | None:
+        return next((a for a in self.artifacts if a.key == key), None)
+
+    def sort_artifacts(self) -> None:
+        """Restore canonical order after parallel generation."""
+        order = {key: i for i, key in enumerate(ARTIFACT_ORDER)}
+        self.artifacts.sort(key=lambda a: order.get(a.key, len(order)))

offerprinter/pricing.py

@@ -0,0 +1,98 @@
+"""Token pricing, so every run can tell you what it actually cost.
+
+The numbers below are **published list prices in USD per 1,000,000 tokens**, and
+they are estimates for display only — providers change prices, and your account
+may have different rates. Nothing here affects generation; it only turns token
+counts into a "this run cost about £0.03" line.
+
+Override any of them in `config.toml`:
+
+    [pricing]
+    "claude-haiku-4-5-20251001" = { input = 1.0, output = 5.0 }
+
+Unknown models fall back to a match on the closest known prefix, and finally to
+zero (in which case OfferPrinter reports tokens but no cost).
+"""
+
+from __future__ import annotations
+
+# model prefix -> (USD per 1M input tokens, USD per 1M output tokens)
+_PRICES: dict[str, tuple[float, float]] = {
+    # --- Anthropic -----------------------------------------------------------
+    "claude-haiku-4-5": (1.00, 5.00),
+    "claude-sonnet-4-5": (3.00, 15.00),
+    "claude-sonnet-4": (3.00, 15.00),
+    "claude-opus-4": (15.00, 75.00),
+    "claude-3-5-haiku": (0.80, 4.00),
+    "claude-3-5-sonnet": (3.00, 15.00),
+    # --- OpenAI --------------------------------------------------------------
+    "gpt-4o-mini": (0.15, 0.60),
+    "gpt-4o": (2.50, 10.00),
+    "gpt-4.1-mini": (0.40, 1.60),
+    "gpt-4.1": (2.00, 8.00),
+    "o4-mini": (1.10, 4.40),
+    # --- Google --------------------------------------------------------------
+    "gemini-1.5-flash": (0.075, 0.30),
+    "gemini-1.5-pro": (1.25, 5.00),
+    "gemini-2.0-flash": (0.10, 0.40),
+    "gemini-2.5-flash": (0.30, 2.50),
+    # --- Moonshot ------------------------------------------------------------
+    "moonshot-v1-8k": (1.68, 1.68),
+    "moonshot-v1-32k": (3.36, 3.36),
+    "moonshot-v1-128k": (8.40, 8.40),
+    "kimi-": (1.68, 1.68),
+}
+
+#: Models that run on your own hardware and therefore cost nothing per token.
+_FREE_PREFIXES = ("llama", "qwen", "mistral", "phi", "gemma", "deepseek", "stub")
+
+#: Rough USD -> GBP rate, only used for the friendly "≈ £0.03" hint.
+USD_TO_GBP = 0.79
+
+
+def rates_for(
+    model: str, overrides: dict[str, tuple[float, float]] | None = None
+) -> tuple[float, float]:
+    """Return (input_price, output_price) per 1M tokens for a model name."""
+    if not model:
+        return (0.0, 0.0)
+    name = model.lower()
+
+    if overrides:
+        if name in overrides:
+            return overrides[name]
+        for prefix, rate in overrides.items():
+            if name.startswith(prefix.lower()):
+                return rate
+
+    if name.startswith(_FREE_PREFIXES):
+        return (0.0, 0.0)
+
+    if name in _PRICES:
+        return _PRICES[name]
+    # Longest matching prefix wins, so "gpt-4o-mini" beats "gpt-4o".
+    matches = [p for p in _PRICES if name.startswith(p)]
+    if matches:
+        return _PRICES[max(matches, key=len)]
+    return (0.0, 0.0)
+
+
+def estimate_cost(
+    model: str,
+    input_tokens: int,
+    output_tokens: int,
+    overrides: dict[str, tuple[float, float]] | None = None,
+) -> float:
+    """Estimated USD cost for a number of input/output tokens."""
+    in_rate, out_rate = rates_for(model, overrides)
+    return (input_tokens * in_rate + output_tokens * out_rate) / 1_000_000
+
+
+def format_cost(usd: float) -> str:
+    """Human-friendly cost string, tiny amounts included."""
+    if usd <= 0:
+        return "free (local model)"
+    gbp = usd * USD_TO_GBP
+    if usd < 0.01:
+        return f"${usd:.4f} (≈ £{gbp:.4f})"
+    return f"${usd:.3f} (≈ £{gbp:.3f})"

offerprinter/prompts/__init__.py

@@ -0,0 +1,32 @@
+"""Prompt templates, kept in their own module so they are easy to audit and improve.
+
+Every generation prompt inherits the same non-negotiable system rules (see
+`SYSTEM_RULES`), the most important of which is: NEVER fabricate. If you want to
+change how OfferPrinter writes, this is the only file you need to touch.
+"""
+
+from offerprinter.prompts.templates import (
+    ATS_REPORT_PROMPT,
+    COVER_LETTER_PROMPT,
+    EXTRACT_META_PROMPT,
+    FIT_MEMO_PROMPT,
+    FIT_SCORE_PROMPT,
+    INTERVIEW_PREP_PROMPT,
+    ROAST_PROMPT,
+    SYSTEM_RULES,
+    TAILORED_CV_PROMPT,
+    build_system,
+)
+
+__all__ = [
+    "SYSTEM_RULES",
+    "build_system",
+    "EXTRACT_META_PROMPT",
+    "TAILORED_CV_PROMPT",
+    "COVER_LETTER_PROMPT",
+    "FIT_MEMO_PROMPT",
+    "ATS_REPORT_PROMPT",
+    "INTERVIEW_PREP_PROMPT",
+    "FIT_SCORE_PROMPT",
+    "ROAST_PROMPT",
+]

offerprinter/prompts/templates.py

@@ -0,0 +1,339 @@
+"""All prompt templates for OfferPrinter.
+
+Design notes:
+- Each artifact has ONE user-prompt template. They all share `SYSTEM_RULES`.
+- Templates use ``{cv}``, ``{jd}``, ``{company}``, ``{role}`` placeholders and are
+  filled with ``str.format`` by the generator service.
+- The no-fabrication rule is repeated in every prompt on purpose. It is the
+  product's core promise and models follow instructions they see close to the task.
+"""
+
+from __future__ import annotations
+
+# ---------------------------------------------------------------------------
+# Shared system rules — the trust guarantee.
+# ---------------------------------------------------------------------------
+
+SYSTEM_RULES = """You are OfferPrinter, an assistant that tailors a person's real \
+job-application materials to a specific role. You write in {locale_name} English \
+({locale_note}).
+
+ABSOLUTE, NON-NEGOTIABLE RULES:
+1. NEVER fabricate. Do not invent experience, skills, employers, job titles, \
+dates, qualifications, certifications, publications, or metrics. You may ONLY \
+reframe, reorder, reword, and surface facts that are actually present in the \
+candidate's CV.
+2. If the role wants something the candidate does not have, DO NOT pretend they \
+have it. Where the artifact calls for honesty about gaps (the fit memo, the ATS \
+report), state the gap plainly. Elsewhere, simply omit what isn't there.
+3. Numbers and metrics must come verbatim from the CV. If the CV says "led a team \
+of 4", never write "a team of 10". If no metric is given, do not invent one.
+4. Keep the candidate's real voice and seniority. Do not inflate titles.
+5. Output clean, ATS-friendly text: no tables, no columns, no images, no text \
+boxes, no fancy characters. Use plain Markdown headings and simple bullet points.
+
+You will be given the candidate's CV and a target job description. Follow the \
+task instructions exactly and return ONLY the requested document in Markdown, \
+with no preamble, no explanation, and no code fences."""
+
+_LOCALE_NOTES = {
+    "UK": ("British", "e.g. 'organise', 'programme', 'CV', 'colour', DD/MM/YYYY dates"),
+    "US": ("American", "e.g. 'organize', 'program', 'resume', 'color', MM/DD/YYYY dates"),
+}
+
+
+def build_system(locale: str) -> str:
+    """Return the shared system prompt localised to UK or US English."""
+    name, note = _LOCALE_NOTES.get(locale, _LOCALE_NOTES["UK"])
+    return SYSTEM_RULES.format(locale_name=name, locale_note=note)
+
+
+# ---------------------------------------------------------------------------
+# Metadata extraction — pull company + role so we can name the output folder.
+# ---------------------------------------------------------------------------
+
+EXTRACT_META_PROMPT = """From the job description below, identify the hiring \
+company's name and the job title.
+
+Return ONLY a single line of the exact form:
+COMPANY: <name> | ROLE: <title>
+
+If the company name is genuinely not stated, use "Company". If the role is not \
+stated, use "Role". Do not add anything else.
+
+--- JOB DESCRIPTION ---
+{jd}
+--- END JOB DESCRIPTION ---"""
+
+
+# ---------------------------------------------------------------------------
+# 1. Tailored CV
+# ---------------------------------------------------------------------------
+
+TAILORED_CV_PROMPT = """Produce a TAILORED CV for this candidate targeting the \
+role of {role} at {company}.
+
+What "tailored" means here:
+- Reorder and reword the candidate's REAL experience so the most role-relevant \
+things come first and are described in the language the job description uses.
+- Foreground the skills, projects and achievements that match this job. \
+De-emphasise (but don't delete truthful, relevant) the rest.
+- Mirror the job's terminology where it truthfully applies to the candidate's \
+experience (helps with ATS keyword matching) — but only where it is genuinely true.
+- Keep every fact, employer, date and metric exactly as in the source CV.
+
+Format (ATS-friendly, plain Markdown):
+# <Candidate Name>
+<contact line: email · phone · location · links — copy from the CV; omit any not present>
+
+## Professional Summary
+A 2–4 sentence summary aimed squarely at this role, built only from real facts.
+
+## Core Skills
+A simple comma or bullet list of the candidate's real skills most relevant to the role.
+
+## Experience
+For each role (most relevant / recent first): title, employer, dates, then 3–6 \
+bullet points rewritten to highlight impact relevant to {role}. Keep real metrics.
+
+## Education
+As in the source CV.
+
+## (Optional) Certifications / Projects
+Only if present in the source CV.
+
+--- CANDIDATE CV ---
+{cv}
+--- END CV ---
+
+--- TARGET JOB DESCRIPTION ---
+{jd}
+--- END JOB DESCRIPTION ---"""
+
+
+# ---------------------------------------------------------------------------
+# 2. Cover letter
+# ---------------------------------------------------------------------------
+
+COVER_LETTER_PROMPT = """Write a TAILORED COVER LETTER for the candidate applying \
+to {role} at {company}.
+
+Requirements:
+- Specific to THIS company and role. No generic filler, no clichés \
+("I am writing to express my interest..."), no flattery that could apply anywhere.
+- Open with a concrete hook tied to the role or company and the candidate's most \
+relevant real strength.
+- 3–4 short paragraphs. In the body, connect the candidate's actual experience to \
+2–3 of the job's key needs, using real examples and metrics from the CV.
+- Confident but honest. Do not claim experience the candidate lacks.
+- Around 250–350 words. Sign off with the candidate's real name.
+
+Format as plain Markdown. Use the candidate's real contact details at the top if \
+present in the CV; otherwise start at the greeting. If you do not know the hiring \
+manager's name, use "Dear Hiring Team,".
+
+--- CANDIDATE CV ---
+{cv}
+--- END CV ---
+
+--- TARGET JOB DESCRIPTION ---
+{jd}
+--- END JOB DESCRIPTION ---"""
+
+
+# ---------------------------------------------------------------------------
+# 3. Fit memo
+# ---------------------------------------------------------------------------
+
+FIT_MEMO_PROMPT = """Write a one-page "WHY I FIT" FIT MEMO for the candidate and \
+the role of {role} at {company}.
+
+This document is where honesty about gaps is REQUIRED. Structure it exactly:
+
+# Fit Memo — {role} at {company}
+
+## Snapshot
+2–3 sentences: the candidate's honest overall fit for this role.
+
+## Requirement-by-requirement
+Identify the key requirements from the job description. For EACH one, write a row \
+in this format:
+- **<the requirement>** — <Strong match | Partial match | Gap>: <one sentence of \
+evidence drawn ONLY from the CV, or a plain statement that the CV does not show it>.
+
+Be scrupulously honest. If the CV shows no evidence for a requirement, mark it \
+**Gap** and say so. Do not invent evidence to fill a gap.
+
+## Honest gaps & how to talk about them
+List the real gaps and, for each, a truthful, non-defensive way the candidate \
+could address it in an interview (transferable experience, willingness to learn, \
+adjacent skills) — WITHOUT claiming they already have the missing skill.
+
+## Bottom line
+One or two sentences on overall suitability.
+
+--- CANDIDATE CV ---
+{cv}
+--- END CV ---
+
+--- TARGET JOB DESCRIPTION ---
+{jd}
+--- END JOB DESCRIPTION ---"""
+
+
+# ---------------------------------------------------------------------------
+# 4. ATS keyword report
+# ---------------------------------------------------------------------------
+
+ATS_REPORT_PROMPT = """Produce an ATS KEYWORD REPORT comparing the candidate's CV \
+against the job description for {role} at {company}.
+
+Steps:
+1. Extract the important keywords and key phrases an Applicant Tracking System \
+would likely screen for from the job description (skills, tools, methods, \
+qualifications, domain terms). Aim for 12–20.
+2. For each, decide whether the candidate's CV already covers it.
+
+Format exactly:
+
+# ATS Keyword Report — {role} at {company}
+
+## Covered keywords
+A bullet list of the job's key terms the CV already contains (quote the CV phrase \
+where helpful).
+
+## Missing keywords
+A bullet list of key terms NOT found in the CV.
+
+## How to add the missing terms — truthfully
+For each missing keyword that the candidate PLAUSIBLY has real experience with \
+(based on the CV), suggest exactly where and how to add it truthfully (e.g. \
+"You list 'built dashboards' — if you used Tableau, name it in that bullet"). \
+For any missing keyword the candidate clearly does NOT have, say plainly: \
+"Do not add — no evidence in your CV. This is a genuine gap." NEVER instruct the \
+candidate to claim something untrue.
+
+## Coverage summary
+"Covered X of Y key terms." with the two numbers.
+
+--- CANDIDATE CV ---
+{cv}
+--- END CV ---
+
+--- TARGET JOB DESCRIPTION ---
+{jd}
+--- END JOB DESCRIPTION ---"""
+
+
+# ---------------------------------------------------------------------------
+# 5. Interview prep pack
+# ---------------------------------------------------------------------------
+
+INTERVIEW_PREP_PROMPT = """Produce an INTERVIEW PREP PACK for the candidate \
+interviewing for {role} at {company}.
+
+Format exactly:
+
+# Interview Prep Pack — {role} at {company}
+
+## Likely questions
+8–10 questions this candidate is likely to be asked for THIS role (mix of \
+behavioural, role-specific/technical, and motivation).
+
+## STAR answer scaffolds
+Pick the 4–5 most important behavioural/experience questions. For each, give a \
+STAR scaffold (Situation, Task, Action, Result) built ONLY from the candidate's \
+real experience in the CV. Where a strong real example exists, use it. Where the \
+CV lacks a directly relevant example, say so honestly and suggest the closest \
+real experience to adapt — do not invent a scenario.
+
+## 5 smart questions to ask them
+Five thoughtful, specific questions the candidate should ask the interviewer \
+about the role, team, or company — showing genuine engagement, not generic.
+
+--- CANDIDATE CV ---
+{cv}
+--- END CV ---
+
+--- TARGET JOB DESCRIPTION ---
+{jd}
+--- END JOB DESCRIPTION ---"""
+
+
+# ---------------------------------------------------------------------------
+# 6. Fit score — a single number, so a run ends with a verdict, not a shrug.
+# ---------------------------------------------------------------------------
+
+FIT_SCORE_PROMPT = """Score how well this candidate genuinely matches the role of \
+{role} at {company}.
+
+Scoring guide (be strict and evidence-based — an inflated score is useless):
+- 85-100: meets essentially every requirement with direct, demonstrated evidence.
+- 70-84: meets most requirements including the critical ones; one or two soft gaps.
+- 55-69: meets roughly half; credible but with real gaps on secondary requirements.
+- 40-54: meets a minority; a genuine stretch with gaps on important requirements.
+- 0-39: missing most requirements, including critical ones.
+
+Judge ONLY on evidence actually present in the CV. Absence of evidence is a gap, \
+not a maybe. Do not be generous to be kind — the candidate needs the truth to \
+decide where to spend their evening.
+
+Return ONLY this exact block, nothing else, no code fences, no commentary:
+
+SCORE: <integer 0-100>
+STRENGTHS: <requirement they genuinely meet> ; <another> ; <another>
+GAPS: <requirement they genuinely do not meet> ; <another>
+
+Give 2-4 strengths and 0-4 gaps, each a short phrase of at most 12 words, \
+separated by semicolons. If there are genuinely no gaps, write "GAPS: none".
+
+--- CANDIDATE CV ---
+{cv}
+--- END CV ---
+
+--- TARGET JOB DESCRIPTION ---
+{jd}
+--- END JOB DESCRIPTION ---"""
+
+
+# ---------------------------------------------------------------------------
+# 7. Roast mode — optional, opt-in, and still never dishonest.
+# ---------------------------------------------------------------------------
+
+ROAST_PROMPT = """Roast this CV. The candidate has explicitly asked for blunt, \
+funny, unsparing feedback — they opted in, so do not soften it into a compliment \
+sandwich.
+
+Rules that still apply (roasting is not licence to be wrong or cruel):
+- Roast the WRITING, never the person, their background, or their circumstances.
+- Every jab must be about something actually in the CV. Do not invent flaws.
+- Punch at clichés, vagueness, buzzwords, unquantified claims, wall-of-text \
+formatting, "responsible for" bullets, and anything that says nothing.
+- Nothing about protected characteristics, age, nationality, or career breaks.
+- End genuinely useful. The point is a better CV, not a worse mood.
+
+Format exactly:
+
+# 🔥 CV Roast
+
+## The verdict in one line
+One brutal but fair sentence.
+
+## What made me wince
+4-6 bullets. Each: quote the offending phrase from the CV, then say why it is \
+doing nothing for them. Be funny. Be specific.
+
+## Cliché count
+List the buzzwords and dead phrases you found, with a count. Crown the worst one.
+
+## Credit where it's due
+2-3 bullets on what genuinely works. Do not invent praise — if something is good, \
+say why it is good.
+
+## The five fixes that would actually change the outcome
+A numbered list of 5 concrete, specific edits, in priority order. Name the exact \
+bullet or section to change and what to change it to.
+
+--- CANDIDATE CV ---
+{cv}
+--- END CV ---"""

offerprinter/services/__init__.py

@@ -0,0 +1,14 @@
+"""Service layer: parsing inputs, generating artifacts, and writing outputs."""
+
+from offerprinter.services.cv_parser import extract_cv, extract_cv_from_bytes
+from offerprinter.services.generator import Generator
+from offerprinter.services.jd_fetcher import load_job_description
+from offerprinter.services.writer import write_package
+
+__all__ = [
+    "extract_cv",
+    "extract_cv_from_bytes",
+    "load_job_description",
+    "Generator",
+    "write_package",
+]