Building a Deterministic JD Analyser: Grounded in the GFE Skill System

Authors: GrowthFlow Engineering Team
Version: 2.0
Last Updated: January 2025
Status: Production

Executive Summary

This whitepaper documents the development journey of the GFE JD Analyser—a deterministic tool that analyses job descriptions and maps them to the GFE Skill System's 174 tasks across Marketing, Sales, and Finance domains.

What began as an LLM-powered experiment evolved into a principled, rule-based system grounded in real task specifications. This document covers:

The Problem: Traditional JD analysis tools either rely on expensive LLMs or use arbitrary "skill" inventories
The Insight: The GFE Skill System already defines 174 concrete tasks—we should match against those, not invent skills
The Architecture: A serverless Netlify function using pattern matching against actual task specifications
The Taxonomy: Complete explanation of the GFE Skill System's structure (Levels, Tasks, Processes, KPIs, ValuationOps)

Part 1: The Problem Space

1.1 Traditional JD Analysis Approaches

Most job description analysis tools fall into two categories:

LLM-Dependent Approaches

Send JD text to GPT/Claude/Gemini and ask it to "extract skills"
Results are non-deterministic and vary between calls
Expensive at scale (API costs compound)
Skills invented on-the-fly with no standardized taxonomy

Keyword-Matching Approaches

Match against arbitrary lists like "Excel", "Communication", "Leadership"
No hierarchy or leveling
No connection to business outcomes
Misses context ("Excel" for a finance role ≠ "Excel" for an admin role)

1.2 The Key Insight

The GFE Skill System already solves this problem. It defines 174 concrete tasks with:

Clear domain assignments (Marketing, Sales, Finance)
Level designations (L0-L4) indicating seniority
KPI linkages connecting tasks to measurable outcomes
Process mappings showing how tasks fit into business workflows

Why invent skills when we have tasks?

A "skill" is abstract and subjective. A task is concrete and verifiable:

Approach	Example	Problem
Invented Skill	"Data Analysis"	What level? What domain? How measured?
GFE Task	`F-012: Prepare Variance Reports` (L1, Finance)	Clear level, domain, linked to KPI-FIN-VAR-01

Part 2: The GFE Skill System Deep Dive

2.1 The Five Levels

The GFE Skill System defines five levels, each with clear progression expectations:

Level	Name	Focus	Time Horizon
L0	Apprentice	Execute defined tasks, build fundamentals	Daily tasks
L1	Practitioner	Own processes, contribute to improvements	Weekly cycles
L2	Architect	Design systems, optimize performance	Monthly/Quarterly
L3	Strategist	Drive cross-functional initiatives	Annual planning
L4	Partner	Shape organizational direction	Multi-year vision

2.2 The LEO Framework

Each level has expectations across three dimensions (Learning, Earning, Org-Building):

┌─────────────────────────────────────────────────────────────┐
│                    LEO Framework                            │
├─────────────────────────────────────────────────────────────┤
│                                                             │
│   Learning (25%)     │   Earning (60%)    │  Org-Building   │
│   ─────────────────  │   ──────────────   │  (15%)          │
│   • Self-study       │   • Task delivery  │  • Mentoring    │
│   • Certifications   │   • KPI impact     │  • Process docs │
│   • Peer learning    │   • Revenue contrib│  • Team growth  │
│                                                             │
└─────────────────────────────────────────────────────────────┘

Each domain (Marketing, Sales, Finance) has specific LEO expectations per level.

2.3 The 174 Tasks

Tasks are the atomic units of the GFE Skill System. Here's the distribution:

Domain	Task Count	ID Prefix	Example
Marketing	67 tasks	M-###	M-001: Clean Contact Lists
Sales	69 tasks	S-###	S-001: Qualify Inbound Leads
Finance	38 tasks	F-###	F-001: Process Invoices

Each task specification includes:

json

{
  "id": "S-015",
  "title": "Negotiate Contract Terms",
  "domain": "sales",
  "level": 2,
  "description": "Lead contract negotiations with prospects...",
  "kpi_links": ["KPI-REV-WIN-01", "KPI-REV-DEAL-01"],
  "owner": "Account Executive",
  "process_id": "PROC-REV-DEAL-CLOSING-01"
}

2.4 The 44 Processes

Tasks group into processes—business workflows with primary KPIs:

Process ID	Process Name	Primary KPI
PROC-REV-DEMAND-GEN-01	Demand Generation Pipeline	KPI-MKT-MQL-01
PROC-REV-DEAL-CLOSING-01	Pipeline Management	KPI-REV-WIN-01
PROC-FIN-REPORTING-01	Financial Reporting	KPI-FIN-REPORT-01
PROC-REV-CUSTOMER-SUCCESS-01	Customer Success Programs	KPI-REV-NRR-01

2.5 ValuationOps Layers

The ultimate purpose: every task connects to company valuation through six layers:

┌────────────────────────────────────────────────────────────────┐
│                   ValuationOps Layers                          │
├────────────────────────────────────────────────────────────────┤
│                                                                │
│   Layer 1: Revenue        → Top-line growth (ARR, MRR)         │
│   Layer 2: EBITDA         → Operational profitability          │
│   Layer 3: Cash           → Working capital, collections       │
│   Layer 4: FCF            → Free cash flow generation          │
│   Layer 5: Story          → Market narrative, positioning      │
│   Layer 6: Valuation      → Multiple expansion, exit value     │
│                                                                │
│   Tasks → Processes → KPIs → ValuationOps Layers               │
│                                                                │
└────────────────────────────────────────────────────────────────┘

Part 3: Technical Architecture

3.1 System Overview

┌─────────────────────────────────────────────────────────────────┐
│                    JD Analyser Architecture                      │
├─────────────────────────────────────────────────────────────────┤
│                                                                  │
│   ┌──────────┐     ┌────────────────┐     ┌────────────────┐    │
│   │  User    │────▶│  VitePress     │────▶│  Netlify       │    │
│   │  Browser │     │  Landing Page  │     │  Function      │    │
│   └──────────┘     └────────────────┘     └────────────────┘    │
│                                                 │                │
│                                                 ▼                │
│                                    ┌────────────────────────┐   │
│                                    │  Embedded Task         │   │
│                                    │  Taxonomy (174 tasks)  │   │
│                                    └────────────────────────┘   │
│                                                 │                │
│                                                 ▼                │
│                                    ┌────────────────────────┐   │
│                                    │  Pattern Matching      │   │
│                                    │  Engine                │   │
│                                    └────────────────────────┘   │
│                                                 │                │
│                                                 ▼                │
│                                    ┌────────────────────────┐   │
│                                    │  Structured Analysis   │   │
│                                    │  Report                │   │
│                                    └────────────────────────┘   │
│                                                                  │
└─────────────────────────────────────────────────────────────────┘

3.2 The Pattern Matching Engine

Instead of LLM inference, we use deterministic keyword pattern matching:

javascript

const TASK_TAXONOMY = {
  "S-001": {
    title: "Qualify Inbound Leads",
    domain: "sales",
    level: 1,
    description: "Review and qualify incoming leads...",
    keywords: ["qualify", "inbound", "lead", "qualification", "mql", "bant"],
    kpi_links: ["KPI-REV-MQL-01"],
    process_id: "PROC-REV-DEMAND-GEN-01"
  },
  // ... 173 more tasks
};

function matchTasksToJD(jdText) {
  const normalizedJD = jdText.toLowerCase();
  const matches = [];
  
  for (const [taskId, task] of Object.entries(TASK_TAXONOMY)) {
    let score = 0;
    let matchedKeywords = [];
    
    for (const keyword of task.keywords) {
      if (normalizedJD.includes(keyword.toLowerCase())) {
        score++;
        matchedKeywords.push(keyword);
      }
    }
    
    if (score > 0) {
      matches.push({
        taskId,
        ...task,
        score,
        matchedKeywords
      });
    }
  }
  
  return matches.sort((a, b) => b.score - a.score);
}

3.3 Why Deterministic > LLM for This Use Case

Aspect	LLM Approach	Deterministic Approach
Cost	$0.01-0.10 per call	$0.00 per call
Consistency	Varies between calls	Same input → same output
Explainability	"The model thinks..."	"Matched keywords: X, Y, Z"
Grounding	Invents skills	Maps to real GFE tasks
Latency	1-5 seconds	<100ms
Offline capable	No	Yes

3.4 Serverless Function Implementation

The Netlify function handles multipart form data (supporting file uploads and text):

javascript

// Key components:
// 1. Busboy for multipart parsing (file + form fields)
// 2. Embedded TASK_TAXONOMY with all 174 tasks
// 3. Pattern matching engine
// 4. Email notification for lead capture (MailerSend)
// 5. Structured JSON response

exports.handler = async (event) => {
  // Parse form data
  const { jdText, company, email, name } = await parseFormData(event);
  
  // Match against taxonomy
  const matchedTasks = matchTasksToJD(jdText);
  
  // Build analysis
  const analysis = buildAnalysis(matchedTasks);
  
  // Send lead notification (non-blocking)
  sendLeadNotification({ company, email, name, matchedTasks });
  
  return {
    statusCode: 200,
    body: JSON.stringify(analysis)
  };
};

Part 4: Analysis Output Structure

4.1 Response Format

The analyser returns a structured response:

json

{
  "summary": {
    "totalMatches": 23,
    "domains": {
      "marketing": 8,
      "sales": 12,
      "finance": 3
    },
    "levels": {
      "L0-Apprentice": 3,
      "L1-Practitioner": 10,
      "L2-Architect": 7,
      "L3-Strategist": 3
    },
    "inferredSeniority": "L1-L2 (Practitioner to Architect)",
    "primaryDomain": "Sales"
  },
  "matchedTasks": [
    {
      "taskId": "S-015",
      "title": "Negotiate Contract Terms",
      "domain": "sales",
      "level": 2,
      "levelName": "Architect",
      "score": 5,
      "matchedKeywords": ["negotiate", "contract", "terms", "pricing", "deal"],
      "kpi_links": ["KPI-REV-WIN-01"],
      "process_id": "PROC-REV-DEAL-CLOSING-01"
    }
    // ... more matches
  ],
  "recommendations": {
    "hiringLevel": "L2 (Architect)",
    "primaryFocus": "Sales Operations with Deal Management",
    "keyProcesses": ["PROC-REV-DEAL-CLOSING-01", "PROC-REV-FORECAST-01"],
    "valuationImpact": "Revenue Layer - Direct impact on win rates and deal velocity"
  }
}

4.2 Level Inference Logic

The system infers seniority based on matched task levels:

javascript

function inferSeniority(matchedTasks) {
  const levelCounts = { 0: 0, 1: 0, 2: 0, 3: 0, 4: 0 };
  
  matchedTasks.forEach(task => {
    levelCounts[task.level]++;
  });
  
  // Weight higher-level tasks more heavily
  const weightedScore = 
    levelCounts[0] * 1 +
    levelCounts[1] * 2 +
    levelCounts[2] * 4 +
    levelCounts[3] * 8 +
    levelCounts[4] * 16;
  
  const avgLevel = weightedScore / matchedTasks.length;
  
  // Map to level range
  if (avgLevel < 1.5) return "L0-L1 (Apprentice to Practitioner)";
  if (avgLevel < 3) return "L1-L2 (Practitioner to Architect)";
  if (avgLevel < 6) return "L2-L3 (Architect to Strategist)";
  return "L3-L4 (Strategist to Partner)";
}

Part 5: Development Journey

5.1 Phase 1: LLM-Based Approach (Failed)

Initial Implementation:

Used Gemini API to analyze JDs
Prompt engineering to extract "skills"
Hit rate limits immediately (429 errors)
Free tier quota exhausted in testing

Lesson Learned: LLMs are expensive for high-volume, deterministic tasks.

5.2 Phase 2: Deterministic with Invented Skills (Wrong)

Second Attempt:

Created pattern matching with arbitrary skill clusters
Used generic level names (Observer, Executor, Expert)
No connection to actual GFE specifications

Lesson Learned: We were reinventing what already existed.

5.3 Phase 3: Grounded in GFE Specs (Correct)

Final Implementation:

Studied actual GFE Skill System specs
Extracted all 174 tasks with metadata
Embedded complete taxonomy in function
Used real level names (Apprentice → Partner)
Connected tasks to processes, KPIs, ValuationOps

Key Insight: Don't abstract—ground in reality.

5.4 Technical Challenges Solved

Challenge	Solution
SSR hydration errors	Client-side only form rendering
Multipart file parsing	Busboy library for Node.js
Email notification failures	Properly awaited async fetch calls
Level name confusion	Direct mapping from spec JSON files
Keyword coverage	Expanded keywords from task descriptions

Part 6: Business Value & Use Cases

6.1 Primary Use Cases

Hiring Managers: Analyze JDs to ensure role clarity and appropriate leveling
Recruiters: Map candidate experience to GFE task competencies
HR Teams: Standardize job architecture across the organization
Candidates: Understand role expectations in GFE taxonomy terms

6.2 Integration Points

The JD Analyser can integrate with:

ATS Systems: Auto-tag roles with GFE task IDs
Performance Management: Link role to expected KPI contributions
Learning Platforms: Generate training paths based on task gaps
Compensation Frameworks: Level-based pay bands

6.3 Lead Generation Value

Every analysis submission captures:

Company name
Contact email
Role type being analyzed
Matched task profile

This enables targeted outreach for GrowthFlow Engineering consulting services.

Part 7: Future Roadmap

7.1 Planned Enhancements

Feature	Status	Impact
PDF resume parsing	Planned	Match candidates to tasks
Batch JD analysis	Planned	Analyze entire job architecture
Gap analysis	Planned	Compare JD to ideal role profile
Process recommendations	In Progress	Suggest missing process coverage
Multilingual support	Planned	Support DE, FR, ES JDs

7.2 Integration Opportunities

GrowthQuiz: Connect to growth mindset assessment
Skill Builder: Generate training recommendations
Team Planner: Analyze team composition against task coverage

Appendix A: Complete Task Reference

Marketing Tasks (67 total)

ID	Title	Level
M-001	Clean Contact Lists	L0
M-002	Schedule Social Posts	L0
M-003	Monitor Campaign Metrics	L1
M-004	Create Landing Pages	L1
M-005	Design Email Templates	L1
...	...	...

Sales Tasks (69 total)

ID	Title	Level
S-001	Qualify Inbound Leads	L1
S-002	Conduct Discovery Calls	L1
S-003	Prepare Proposals	L1
S-015	Negotiate Contract Terms	L2
S-030	Manage Key Accounts	L2
...	...	...

Finance Tasks (38 total)

ID	Title	Level
F-001	Process Invoices	L0
F-002	Reconcile Accounts	L1
F-012	Prepare Variance Reports	L1
F-025	Build Financial Models	L2
F-035	Lead Budget Planning	L3
...	...	...

Full task specifications available in the GFE Skill System repository.

Appendix B: API Reference

Endpoint

POST https://growthflowengineering.xyz/.netlify/functions/jd-analyser

Request (multipart/form-data)

Field	Type	Required	Description
jdText	string	Yes	Job description text
company	string	Yes	Company name
email	string	Yes	Contact email
name	string	No	Contact name
jdFile	file	No	JD file (PDF/DOCX)

Response

json

{
  "success": true,
  "analysis": {
    "summary": { ... },
    "matchedTasks": [ ... ],
    "recommendations": { ... }
  }
}

Conclusion

The GFE JD Analyser demonstrates a fundamental principle: grounded systems beat generic AI.

By embedding the actual GFE Skill System taxonomy—174 tasks, 44 processes, 5 levels—we created a tool that:

Costs nothing to run (no LLM API calls)
Produces consistent results (deterministic matching)
Explains its reasoning (shows matched keywords)
Connects to business value (tasks → processes → KPIs → ValuationOps)

This is the future of domain-specific AI: systems grounded in real taxonomies, not black-box models inventing skills on the fly.

Try the JD Analyser

Analyse Your Job Description →

Get instant insights into role requirements mapped to the GFE Skill System.

For questions or partnership inquiries, contact the GrowthFlow Engineering team.

Building a Deterministic JD Analyser: Grounded in the GFE Skill System ​

Executive Summary ​

Part 1: The Problem Space ​

1.1 Traditional JD Analysis Approaches ​

1.2 The Key Insight ​

Part 2: The GFE Skill System Deep Dive ​

2.1 The Five Levels ​

2.2 The LEO Framework ​

2.3 The 174 Tasks ​

2.4 The 44 Processes ​

2.5 ValuationOps Layers ​

Part 3: Technical Architecture ​

3.1 System Overview ​

3.2 The Pattern Matching Engine ​

3.3 Why Deterministic > LLM for This Use Case ​

3.4 Serverless Function Implementation ​

Part 4: Analysis Output Structure ​

4.1 Response Format ​

4.2 Level Inference Logic ​

Part 5: Development Journey ​

5.1 Phase 1: LLM-Based Approach (Failed) ​

5.2 Phase 2: Deterministic with Invented Skills (Wrong) ​

5.3 Phase 3: Grounded in GFE Specs (Correct) ​

5.4 Technical Challenges Solved ​

Part 6: Business Value & Use Cases ​

6.1 Primary Use Cases ​

6.2 Integration Points ​

6.3 Lead Generation Value ​

Part 7: Future Roadmap ​

7.1 Planned Enhancements ​

7.2 Integration Opportunities ​

Appendix A: Complete Task Reference ​

Marketing Tasks (67 total) ​

Sales Tasks (69 total) ​

Finance Tasks (38 total) ​

Appendix B: API Reference ​

Endpoint ​

Request (multipart/form-data) ​

Response ​

Conclusion ​

Try the JD Analyser ​

Building a Deterministic JD Analyser: Grounded in the GFE Skill System

Executive Summary

Part 1: The Problem Space

1.1 Traditional JD Analysis Approaches

1.2 The Key Insight

Part 2: The GFE Skill System Deep Dive

2.1 The Five Levels

2.2 The LEO Framework

2.3 The 174 Tasks

2.4 The 44 Processes

2.5 ValuationOps Layers

Part 3: Technical Architecture

3.1 System Overview

3.2 The Pattern Matching Engine

3.3 Why Deterministic > LLM for This Use Case

3.4 Serverless Function Implementation

Part 4: Analysis Output Structure

4.1 Response Format

4.2 Level Inference Logic

Part 5: Development Journey

5.1 Phase 1: LLM-Based Approach (Failed)

5.2 Phase 2: Deterministic with Invented Skills (Wrong)

5.3 Phase 3: Grounded in GFE Specs (Correct)

5.4 Technical Challenges Solved

Part 6: Business Value & Use Cases

6.1 Primary Use Cases

6.2 Integration Points

6.3 Lead Generation Value

Part 7: Future Roadmap

7.1 Planned Enhancements

7.2 Integration Opportunities

Appendix A: Complete Task Reference

Marketing Tasks (67 total)

Sales Tasks (69 total)

Finance Tasks (38 total)

Appendix B: API Reference

Endpoint

Request (multipart/form-data)

Response

Conclusion

Try the JD Analyser