Resume Humanizer

When to Use This Skill

Use this skill as Phase 3 — after bullet generation (PACTI), before LaTeX formatting. Run it on all bullets before they enter the LaTeX template.

Trigger phrases: "humanize my resume", "remove AI language", "make this sound more natural", "style check", "review my bullets"

Also use proactively when you detect:

• Any of the flagged words (leveraged, spearheaded, pioneered, etc.) appearing in bullets
• Bullets that read like marketing copy rather than engineering notes
• Bullets with no technology or system name
• Bullets with adjectives but no concrete nouns

---

Core Principle

This is a style checker, not a word filter. Words are not banned. They are penalized when overused or when they substitute for specific technical content. "Leveraged Kubernetes autoscaling" is perfectly natural. "Leveraged cutting-edge AI to deliver innovative solutions" is not. The difference is specificity, not the word itself.

---

7-Dimension Scoring System

Score the full set of resume bullets on each dimension (0–10). All dimensions must meet their target before the resume proceeds to LaTeX generation.

#	Dimension	Target Score	What's Measured
1	AI Buzzword Score	≤2/10	Frequency and clustering of AI-cliché words
2	Specificity Score	≥8/10	Every bullet names ≥1 technology, algorithm, system, or proper noun
3	Evidence Score	10/10	Every bullet traceable to the evidence database or GitHub source
4	Metric Density	≥5/10	Measurable outcomes preferred; concrete technical implementation details (named technologies, specific endpoints, component counts) count toward this score when real metrics aren't available
5	Technical Density	≥6/10	Architecture / algorithm / design reasoning present in technical bullets. PACTI "Core Technical Decision" is recommended but not mandatory — direct implementation statements with named technologies also score well
6	Readability	≥8/10	Max 1 adjective per bullet; no filler words; ≤2 lines when rendered
7	Interview Defensibility	10/10	Every bullet has a Q&A pair in the evidence database that the candidate can answer

---

Dimension 1: AI Buzzword Score (target ≤2/10)

Flagged Words — Penalized When Overused

These words are not banned. They are penalized if:

• Used more than once across the full resume, OR
• Used as a substitute for specific technical content (i.e., the word is the only content)

Word	Acceptable Use	Flagged Use
leveraged	"Leveraged Kubernetes autoscaling to handle traffic spikes"	"Leveraged cutting-edge AI techniques"
spearheaded	"Spearheaded migration to microservices (once, specific)"	Used 3× across resume
pioneered	"Pioneered adoption of Rust in the team's backend codebase"	"Pioneered innovative solutions"
orchestrated	"Orchestrated LangGraph multi-agent pipeline"	"Orchestrated seamless collaboration"
championed	Used once, describing specific advocacy	Used as vague leadership claim
utilized	Almost always replaceable with "used"	"Utilized various technologies"
seamless	Should never appear — always vague	Any usage
cutting-edge	Should never appear — always vague	Any usage
innovative	Should never appear — claims are hollow	Any usage
robust	Should never appear — show it, don't say it	Any usage
passionate	Should never appear in technical bullets	Any usage
dynamic	Should never appear — always vague	Any usage
results-driven	Should never appear — should be evident	Any usage
synergy	Should never appear	Any usage
state-of-the-art	Should never appear — use specific benchmark	Any usage

Preferred Replacements

Instead of	Use
leveraged, utilized	built, used, applied
spearheaded	led, initiated, designed
pioneered	introduced, created, first to implement
orchestrated	coordinated, built, structured
championed	advocated for, proposed, implemented
seamless	(delete — describe what was smooth instead)
cutting-edge	name the specific technology
innovative	describe what was novel specifically

---

Dimension 2: Specificity Score (target ≥8/10)

Every bullet must contain at least one of:

• A named technology (FastAPI, LangGraph, Isolation Forest, PostgreSQL, React)
• A named algorithm or model (Isolation Forest, LSTM, BFS, Transformer)
• A named system or service (GitHub Actions, Celery, Redis, Kafka)
• A named design pattern (event-driven, microservice, CQRS)
• A specific metric with units (40%, 3 minutes, 500ms, 100K requests)

Specificity check — rewrite triggers:

❌ "Built a system to handle large amounts of data efficiently"
✅ "Built a batch processing pipeline with Apache Kafka consuming 50K events/hour"

❌ "Implemented machine learning for anomaly detection"
✅ "Implemented Isolation Forest anomaly detection on unlabeled infrastructure logs"

❌ "Improved system performance significantly"
✅ "Reduced API response time from 850ms to 210ms by adding Redis caching"

---

Dimension 5: Technical Density (target ≥6/10)

Technical bullets should ideally contain engineering reasoning, not just facts. The PACTI structure helps with this — specifically the "Core Technical Decision" step — but is NOT mandatory for every bullet.

Low technical density (rewrite):

Built an anomaly detection system using Isolation Forest.

High technical density (keep — PACTI style):

Chose Isolation Forest over DBSCAN for anomaly detection because the log data lacked labels; the unsupervised nature of IF made it robust to varying log densities across microservices.

Acceptable technical density (keep — direct style):

Built a REST API with FastAPI serving 4 endpoints for log ingestion and retrieval, with structured error handling and request validation using Pydantic models.

The second example uses PACTI and explains why. The third example doesn't use PACTI but still has high specificity (named technology, component count, specific patterns). Both are acceptable. Not every bullet needs to be a mini architecture decision record.

---

Dimension 6: Readability (target ≥8/10)

Filler Word Removal

Delete on sight:

• "successfully" (if you did it, it was successful)
• "effectively" (same logic)
• "various" (name them)
• "multiple" (count them)
• "many" (count or estimate: "10+", "~50")
• "several" (same)
• "cutting-edge", "state-of-the-art" (name the tech)

Adjective Budget

Maximum 1 adjective per bullet. Adjectives that add specificity are fine ("asynchronous", "distributed", "real-time"). Adjectives that are opinions are not ("innovative", "robust", "efficient" without a metric).

Line Length

Maximum 2 lines when rendered in LaTeX. If a bullet is longer:

1. Split into two bullets (each must still meet all 7 dimensions independently)
2. Or cut to the most impactful single fact + metric

---

Writing Style Application

The style profile from the JD Intelligence Briefing is applied here. The tone, verb choices, and emphasis shift based on the 2D matrix output.

Profile	Tone Guidance
Startup/OS × Eng Manager	Ownership language: "designed", "built end-to-end", "shipped". Architecture reasoning required.
Enterprise × Eng Manager	Reliability and scale: SLA, fault tolerance, distributed system context.
Research × Researcher	Methodology and rigor: experiment design, datasets used, evaluation metrics
DevTools × Recruiter	Developer experience: API design quality, adoption metrics, documentation
Big Tech × Eng Manager	Scale and impact: users served, requests/sec, latency, infra cost
Open Source × Recruiter	Contribution clarity: PRs merged, stars, community impact

---

Humanizer Pass Output Format

Run the 7-dimension scorecard on the full bullet set. Output:

## Humanizer Report

### Scores
| Dimension | Score | Status |
|---|---|---|
| AI Buzzword | 1/10 | ✅ Pass |
| Specificity | 9/10 | ✅ Pass |
| Evidence | 10/10 | ✅ Pass |
| Metric Density | 7/10 | ✅ Pass |
| Technical Density | 8/10 | ✅ Pass |
| Readability | 9/10 | ✅ Pass |
| Interview Defensibility | 10/10 | ✅ Pass |

### Applied Style Profile
[Startup/OS × Eng Manager]

### Flags
- Bullet 3: "utilized various techniques" → No specific technology named. Rewrite required.
- Bullet 7: "spearheaded" used twice across resume. Second usage flagged. Rewrite one.

### Rewrites
**Original:** "Utilized various NLP techniques for text processing"
**Revised:** "Applied spaCy for tokenization and entity extraction on 50K Hinglish support tickets"

**Original (Bullet 7 second use):** "Spearheaded data pipeline improvements"
**Revised:** "Redesigned the ingestion pipeline to process logs asynchronously, reducing backlog from 4 hours to under 10 minutes"

---

Rules

1. Style rules are checks, not blocks. Flag and suggest; never silently delete content.
2. Writing style comes from the JD Intelligence Briefing. If no briefing exists, default to [Enterprise × Eng Manager].
3. All 7 dimensions must pass before proceeding to LaTeX. If any fail, rewrite and re-score.
4. Do not invent specificity. If a bullet lacks a technology name and the evidence database has no source for one, flag it — do not fabricate a technology name.
5. Readability beats cleverness. A simple, precise bullet beats a sophisticated but confusing one.