CS + Engineering Resume Dataset Generator
An n8n workflow that generates demographically diverse, photorealistic synthetic resume datasets — LLM-written content, ComfyUI headshots, Reactive Resume PDFs, Drive delivery.
Repo: https://github.com/jerohalili/resume-ai-generator-automation Self-hosted pipeline — no public demo; runs via Docker (n8n + AnythingLLM + ComfyUI).
The Challenge
HR-tech needs large, diverse resume datasets to train ATS parsers and NER models — but real resumes carry GDPR/NPC privacy risk, and hand-writing thousands of consistent profiles is impossible.
The Solution
I built a “set and forget” n8n workflow that generates 2 candidate profiles per run end-to-end: randomized diverse personas → two LLM passes for resume JSON → ComfyUI headshot → Reactive Resume create/patch/PDF → Google Drive upload (JSON + PDF + headshot). A GitHub Actions schedule can trigger dataset builds automatically.
Technical Highlights
Procedural Persona Engine
- Diversity by design: Weighted ethnic pools with culturally matched names plus ComfyUI prompt descriptors; random M/F assignment; specialties across CS (ML/AI, Backend, Frontend, DevOps, Security, Mobile) and Engineering (Mechanical, Civil, Electrical, Chemical, Aerospace, Biomedical).
- Consistency: Career progression, skill clusters, and visual identity stay aligned per persona.
Full Pipeline Nodes
Define Profiles → LLM Call 1 (basics/experience/education) → LLM Call 2 (skills/projects/certs) → Assemble → ComfyUI headshot (150s wait) → Drive upload → Reactive Resume create + patch → 15s printer wait → PDF export → Drive upload (PDF + JSON).
Scheduled + Credentialed
- GitHub Actions (
.github/workflows/generate_dataset.yml) fires n8n via REST API — needsN8N_BASE_URL,N8N_API_KEY,N8N_WORKFLOW_IDsecrets; expose local n8n via Cloudflare Tunnel for remote triggers. - n8n credentials: Reactive Resume (
x-api-keyheader), AnythingLLM (Bearerheader), Google Drive (OAuth2 client ID/secret).
System Architecture
| Layer | Technology | Role |
|---|---|---|
| Orchestration | n8n | Imports workflow/CS+Engineering Resume Dataset Generator v4.json. |
| Text Generation | AnythingLLM (Llama 3.1 GGUF) | Two-pass resume JSON via chat API. |
| Image Synthesis | ComfyUI (Stable Diffusion) | Photorealistic headshots per persona. |
| Document Engine | Reactive Resume v5 API | Resume create, patch, and PDF export. |
| Storage | Google Drive API | JSON + PDF + headshot per candidate. |
Impact & Key Takeaways
Turns a week of data entry into a background job: 10 or 1,000 resumes cost the same effort, with zero PII risk and dual outputs (JSON for model training, PDF for parser/OCR testing). Lesson: API-first document engines turn any structured data into polished artifacts.
Roadmap
- Short-Term: Multi-language generation (Tagalog/English/Spanish).
- Long-Term: LinkedIn-profile generator for holistic synthetic footprints.
How to Run This Project
git clone https://github.com/jerohalili/resume-ai-generator-automation.git
cd resume-ai-generator-automation
chmod +x scripts/setup.sh
./scripts/setup.sh
# fill .env (ANYTHINGLLM_API_KEY, ANYTHINGLLM_WORKSPACE,
# COMFYUI_MODEL, REACTIVE_RESUME_API_KEY,
# GDRIVE_CLIENT_ID/SECRET, GDRIVE_FOLDER_ID)
./scripts/setup.sh
chmod +x scripts/import_workflow.sh
./scripts/import_workflow.sh
# open http://localhost:5678, configure credentials, hit Execute WorkflowPrerequisites: Docker + Compose v2, NVIDIA GPU recommended, Reactive Resume account + API key, Google Cloud project with Drive API enabled, GGUF model in AnythingLLM, SD checkpoint in ComfyUI.