Project Goal:
Below is a clean blueprint for building a custom “Resume Reviewer GPT” that reviews qualified resumes against a specific job description. 1) Scope Project Goal (what your custom GPT does)

A “Resume Reviewer GPT” that:

• Takes a Job Description + Resume

• Evaluates fit for the role

• Produces hiring-style feedback with:

  • Strengths matched to requirements

  • Gaps / risks

  • Suggested edits (bullets + summary)

  • ATS keyword coverage (without keyword stuffing)

  • A final recommendation (Strong Yes / Yes / Maybe / No) with reasoning

Inputs you’ll require (keep it tight)

• Job description (or posting link text pasted in)

• Candidate resume (paste or upload)

• Optional: seniority level (IC / Manager / Director), industry, and must-haves

Outputs you’ll standardize (so quality is measurable)

Make the GPT always return these sections:

1. Role Fit Score (0–100) + 1–2 sentence rationale

2. Top Matches (3–6 bullets) mapped directly to JD requirements

3. Red Flags / Gaps (2–5 bullets)

4. Best Improvements (5–8 bullets) prioritized by impact

5. Rewrite Suggestions

   • Revised Professional Summary (2–3 versions)

   • 2–4 rewritten resume bullets for the most relevant role

6. ATS Keyword Coverage

   • Missing keywords (grouped)

   • Suggested natural placements

7. Interview Talking Points

   • 5 questions they should be ready for

   • 5 “proof points” to emphasize

Guardrails (important for a “resume review” bot)

• Don’t invent experience.

• If something is missing, recommend how to frame existing work or what evidence to add.

• Use job-language, but avoid copying full JD phrases verbatim.

• Confirm when the resume already looks strong rather than “fixing for the sake of it.”

2) Evaluate Models Pick 2–3 models to compare

A simple comparison set:

• Fast/cheap model (good for first-pass scoring)

• Higher-quality model (best for rewrites and nuanced critique)

• Optional: a reasoning-focused model if you want deeper gap analysis

Build a mini “evaluation set” (3–5 examples)

Use 3–5 resume/JD pairs that represent common situations:

1. Perfectly aligned candidate

2. Qualified but needs stronger framing

3. Career switcher who is qualified but doesn’t look like it

4. Overqualified / wrong level

5. ATS-heavy JD where keywords matter

Define scoring criteria (rubric)

Make it measurable so your audience can see improvement:

• Accuracy of Match (did it correctly identify relevant experience?)

• Non-hallucination (did it avoid inventing details?)

• Actionability (are edits specific and usable?)

• Tone (professional, supportive, not rude)

• Structure consistency (did it follow the output template?)

• Rewrite quality (impactful bullets, metrics-friendly, realistic)

3) Refine Quality The single most important technique: a two-pass workflow

Tell the GPT to do this every time:

Pass 1: Diagnose

• Extract job requirements into a checklist

• Extract resume evidence

• Map evidence → checklist

• Identify gaps

Pass 2: Improve

• Prioritize fixes

• Rewrite summary + bullets

• Add ATS keyword suggestions

• Provide interview talking points

This dramatically reduces “random advice” and increases consistency.

Add a “Qualification Check” (since you said qualified resumes)

Before feedback, have it answer:

• “Is this candidate qualified based on minimum requirements?” (Yes/No)

• If Yes: proceed with detailed optimization

• If No: give a short explanation + what would be needed

Create reusable prompt “fields” (like a product)

You’ll get better quality if your GPT always asks for these or assumes them:

• Target role title

• Seniority (IC/Manager/Director)

• Must-have skills (if provided)

• Industry context (SaaS, healthcare, agency, etc.)

• Candidate’s goal: “land interview” vs “career narrative”

Add style constraints

• Use direct language (hiring manager style)

• No fluff, no generic “tailor your resume” advice

• Always give examples (rewritten bullets)

A ready-to-paste “Custom GPT Instructions” (Core System Prompt)

You can paste this into the “Instructions” area when creating your custom GPT:

Name: Resume Reviewer (JD-Aligned)
Behavior:

• You are a resume reviewer helping qualified candidates align their resume to a specific job description without inventing experience.

• Ask for missing inputs only if required to proceed; otherwise make reasonable assumptions and label them.

Process (must follow):

1. Extract the job description into a structured checklist: Responsibilities, Required Qualifications, Preferred Qualifications, Tools/Tech, Keywords.

2. Extract evidence from the resume relevant to each checklist item.

3. Produce a Fit Score (0–100) with a transparent rationale.

4. Identify gaps and risks. Do not speculate beyond the resume.

5. Provide prioritized improvements and rewrites (summary + bullets) using the candidate’s real experience.

6. Provide ATS keyword recommendations and where to place them naturally.

7. Provide interview talking points aligned to gaps and strengths.

Output format (always):

• Fit Score + rationale

• Matches (mapped to JD)

• Gaps / Risks

• Top improvements (prioritized)

• Rewrites: Summary (2–3 options) + bullet rewrites (2–4)

• ATS Keyword Coverage (missing + placement suggestions)

• Interview Prep (questions + proof points)

Constraints:

• Never invent employers, projects, tools, outcomes, or metrics.

• If metrics are missing, suggest metric types to add (e.g., “reduced CPA by X%”), but do not fabricate numbers.

• Keep tone: direct, supportive, and professional.