feat: add interview prep workflow

README.md

@@ -50,7 +50,7 @@ Resume Matcher works by creating a master resume that you can use to tailor for
 1. **Upload** your master resume (PDF or DOCX)
 2. **Paste** a job description you're targeting
 3. **Review** AI-generated improvements and tailored content
-4. **Cover Letter** generator for the job application
+4. **Cover Letter** and optional interview preparation for the job application
 5. **Customize** the layout and sections to fit your style
 6. **Export** as a professional PDF with your preferred template
 
@@ -145,6 +145,10 @@ Generate tailored cover letters based on the job description and your resume.
 
 ![Cover Letter](assets/cover_letter.png)
 
+### Interview Preparation
+
+Generate structured, resume-grounded interview prep for saved tailored resumes. Use the Builder's Interview Prep tab on demand, or enable automatic generation in Settings.
+
 ### Resume Scoring & Keyword Highlighting
 
 Analyze your resume against the job description with a match score, keyword highlighting, and suggestions for improvement.

apps/backend/app/database.py

@@ -126,6 +126,7 @@ def _resume_to_dict(row: Resume) -> dict[str, Any]:
             "processing_status": row.processing_status,
             "cover_letter": row.cover_letter,
             "outreach_message": row.outreach_message,
+            "interview_prep": row.interview_prep,
             "title": row.title,
             "created_at": row.created_at,
             "updated_at": row.updated_at,
@@ -191,6 +192,7 @@ async def create_resume(
         outreach_message: str | None = None,
         title: str | None = None,
         original_markdown: str | None = None,
+        interview_prep: str | None = None,
     ) -> dict[str, Any]:
         """Create a new resume entry.
 
@@ -211,6 +213,7 @@ async def create_resume(
                     processing_status=processing_status,
                     cover_letter=cover_letter,
                     outreach_message=outreach_message,
+                    interview_prep=interview_prep,
                     title=title,
                     original_markdown=original_markdown,
                     created_at=now,
@@ -230,6 +233,7 @@ async def create_resume(
             "processing_status": processing_status,
             "cover_letter": cover_letter,
             "outreach_message": outreach_message,
+            "interview_prep": interview_prep,
             "title": title,
             "created_at": now,
             "updated_at": now,
@@ -249,6 +253,7 @@ async def create_resume_atomic_master(
         outreach_message: str | None = None,
         original_markdown: str | None = None,
         title: str | None = None,
+        interview_prep: str | None = None,
     ) -> dict[str, Any]:
         """Create a new resume with atomic master assignment.
 
@@ -281,6 +286,7 @@ async def create_resume_atomic_master(
                 processing_status=processing_status,
                 cover_letter=cover_letter,
                 outreach_message=outreach_message,
+                interview_prep=interview_prep,
                 original_markdown=original_markdown,
                 title=title,
             )

apps/backend/app/db_engine.py

@@ -62,3 +62,10 @@ def make_sync_engine(path: Path) -> Engine:
 def init_models_sync(engine: Engine) -> None:
     """Create all tables (idempotent) using a sync engine connection."""
     Base.metadata.create_all(engine)
+
+    # ``create_all`` does not ALTER existing SQLite tables. Keep this additive
+    # migration idempotent so older local databases can load resumes safely.
+    with engine.begin() as conn:
+        columns = conn.exec_driver_sql("PRAGMA table_info(resumes)").mappings().all()
+        if columns and "interview_prep" not in {column["name"] for column in columns}:
+            conn.exec_driver_sql("ALTER TABLE resumes ADD COLUMN interview_prep TEXT")

apps/backend/app/llm.py

@@ -814,7 +814,8 @@ def _appears_truncated(data: dict, schema_type: str = "resume") -> bool:
     Args:
         data: Parsed JSON dict.
         schema_type: Expected schema — "resume" (full resume), "enrichment"
-            (analyze output), "diff" (diff changes), or "keywords".
+            (analyze output), "diff" (diff changes), "keywords", or
+            "interview_prep".
             Determines which fields are checked for truncation.
     """
     if not isinstance(data, dict):
@@ -844,6 +845,23 @@ def _appears_truncated(data: dict, schema_type: str = "resume") -> bool:
             return True
         return False
 
+    if schema_type == "interview_prep":
+        required = {
+            "role_fit_analysis",
+            "resume_questions",
+            "project_follow_ups",
+            "skill_gaps",
+            "talking_points",
+        }
+        missing = required - set(data)
+        if missing:
+            logging.warning(
+                "Possible truncation detected: interview_prep missing required keys: %s",
+                ", ".join(sorted(missing)),
+            )
+            return True
+        return False
+
     # For "diff", "keywords", and unknown schemas: no truncation heuristics.
     # Diff may legitimately return empty changes; keywords may return empty
     # lists when the job description has no actionable terms.
@@ -1064,9 +1082,9 @@ async def complete_json(
     are handled by the Router and are NOT retried again here.
 
     Args:
-        schema_type: Expected schema — "resume", "enrichment", "diff", or
-            "keywords". Passed to _appears_truncated for context-aware truncation
-            detection and used to tailor retry hints.
+        schema_type: Expected schema — "resume", "enrichment", "diff",
+            "keywords", or "interview_prep". Passed to _appears_truncated for
+            context-aware truncation detection and used to tailor retry hints.
     """
     router, config = get_router(config)
     model_name = get_model_name(config)
@@ -1134,6 +1152,10 @@ async def complete_json(
                         hint = (
                             "\n\nIMPORTANT: Output the COMPLETE JSON object with ALL keys: items_to_enrich, questions, analysis_summary. Do not truncate."
                         )
+                    elif schema_type == "interview_prep":
+                        hint = (
+                            "\n\nIMPORTANT: Output the COMPLETE JSON object with ALL keys: role_fit_analysis, resume_questions, project_follow_ups, skill_gaps, talking_points. Do not truncate."
+                        )
                     else:
                         hint = (
                             "\n\nIMPORTANT: Output ONLY a valid JSON object. Start with { and end with }."

apps/backend/app/models.py

@@ -42,6 +42,7 @@ class Resume(Base):
     processing_status: Mapped[str] = mapped_column(String, default="pending")
     cover_letter: Mapped[str | None] = mapped_column(Text, nullable=True)
     outreach_message: Mapped[str | None] = mapped_column(Text, nullable=True)
+    interview_prep: Mapped[str | None] = mapped_column(Text, nullable=True)
     title: Mapped[str | None] = mapped_column(String, nullable=True)
     # original_markdown has *absence* semantics in the TinyDB era: the key was
     # omitted entirely when None. The facade reproduces that by only emitting

apps/backend/app/prompts/__init__.py

@@ -10,6 +10,7 @@
     IMPROVE_PROMPT_OPTIONS,
     IMPROVE_RESUME_PROMPT,
     IMPROVE_RESUME_PROMPTS,
+    INTERVIEW_PREP_PROMPT,
     PARSE_RESUME_PROMPT,
     SKILL_TARGET_PLAN_PROMPT,
     get_language_name,
@@ -51,6 +52,7 @@ def validate_prompt_placeholders(prompt: str) -> list[str]:
     "DIFF_STRATEGY_INSTRUCTIONS",
     "SKILL_TARGET_PLAN_PROMPT",
     "GENERATE_TITLE_PROMPT",
+    "INTERVIEW_PREP_PROMPT",
     "REQUIRED_FEATURE_PROMPT_PLACEHOLDERS",
     "validate_prompt_placeholders",
     "get_language_name",

apps/backend/app/prompts/templates.py

@@ -396,6 +396,60 @@ def _build_truthfulness_rules(rule_7: str) -> str:
 
 Output plain text only. No JSON, no markdown formatting."""
 
+INTERVIEW_PREP_PROMPT = """Generate structured interview preparation for this tailored resume and job.
+
+IMPORTANT: Write in {output_language}.
+Do NOT translate JSON property names. Keep every JSON key exactly as shown in the schema; translate only string values.
+
+Job Description:
+{job_description}
+
+Candidate Resume (JSON):
+{resume_data}
+
+Truthfulness guardrails:
+- Use only evidence from the resume JSON and job description.
+- Do NOT invent experience, tools, employers, metrics, certifications, skills, responsibilities, education, projects, or claims beyond the provided evidence.
+- Do NOT imply the candidate has a skill or background unless it is present in the resume.
+- Skill gaps are preparation targets only. They are not claimed candidate skills.
+- If a job requirement is not evidenced by the resume, present it as something to prepare for or explain honestly.
+
+Return ONLY a valid JSON object with exactly these top-level keys:
+{{
+  "role_fit_analysis": ["Short evidence-based role-fit observation"],
+  "resume_questions": [
+    {{
+      "question": "Interview question grounded in the resume and job",
+      "focus_area": "Resume evidence or job requirement being tested",
+      "suggested_answer_points": ["Truthful point based on resume evidence"]
+    }}
+  ],
+  "project_follow_ups": [
+    {{
+      "question": "Follow-up question about a real resume project or experience",
+      "focus_area": "Project, impact, tradeoff, or implementation detail",
+      "suggested_answer_points": ["Truthful point based on resume evidence"]
+    }}
+  ],
+  "skill_gaps": [
+    {{
+      "skill": "Job-relevant skill or topic to prepare",
+      "why_it_matters": "Why this topic may come up for this role",
+      "preparation_suggestion": "How to prepare without claiming unsupported experience"
+    }}
+  ],
+  "talking_points": ["Concise role-specific talking point grounded in the resume"]
+}}
+
+Content requirements:
+- role_fit_analysis: 3-5 bullets.
+- resume_questions: 5-8 questions.
+- project_follow_ups: 3-6 questions.
+- skill_gaps: 3-5 preparation targets.
+- talking_points: 5-8 concise points.
+- Keep all suggested answer points factual and resume-grounded.
+- Do NOT use markdown fences or commentary outside the JSON."""
+
 GENERATE_TITLE_PROMPT = """Extract the job title and company name from this job description.
 
 IMPORTANT: Write in {output_language}.

apps/backend/app/routers/config.py

@@ -225,6 +225,7 @@ async def get_feature_config() -> FeatureConfigResponse:
     return FeatureConfigResponse(
         enable_cover_letter=stored.get("enable_cover_letter", False),
         enable_outreach_message=stored.get("enable_outreach_message", False),
+        enable_interview_prep=stored.get("enable_interview_prep", False),
     )
 
 
@@ -238,13 +239,16 @@ async def update_feature_config(request: FeatureConfigRequest) -> FeatureConfigR
         stored["enable_cover_letter"] = request.enable_cover_letter
     if request.enable_outreach_message is not None:
         stored["enable_outreach_message"] = request.enable_outreach_message
+    if request.enable_interview_prep is not None:
+        stored["enable_interview_prep"] = request.enable_interview_prep
 
     # Save config
     _save_config(stored)
 
     return FeatureConfigResponse(
         enable_cover_letter=stored.get("enable_cover_letter", False),
         enable_outreach_message=stored.get("enable_outreach_message", False),
+        enable_interview_prep=stored.get("enable_interview_prep", False),
     )