The n8n workflow runs as a sequential agentic pipeline with 13 nodes.

Node-by-Node Breakdown

Node 1 — Webhook Trigger

Method: POST. Receives jd_text as plain text and resume_pdf as base64-encoded PDF data. Configured for persistent uptime via Cloudflare Zero Trust tunnel on Oracle Cloud VM.

Node 2 — API Key Validation

Code node. Pass-through validation layer. Confirms the incoming payload is properly formed before any processing begins.

Node 3 — Input Validation

Code node. Extracts jd_text and resume_pdf from the request body. Validates both fields exist. Confirms job description is at least 50 characters. Structures the payload for downstream nodes.

Node 4 — Start Latency Timer

Code node. Records the current timestamp in milliseconds. Used at the end of the pipeline to calculate total evaluation latency.

Node 5 — Validate File

Code node. Strips the base64 data URL prefix if present. Converts the base64 string to a binary buffer. Outputs the binary PDF data in the format the Extract PDF Text node requires.

Node 6 — Extract PDF Text

n8n built-in Extract from File node. Operation: Extract From PDF. Input Binary Field: data. Extracts all readable text from the PDF regardless of layout, column structure, or formatting.

Node 7 — Check Extraction Success

IF node. Condition: extracted text field is not empty and length is at least 100 characters. TRUE path continues to evaluation. FALSE path returns a clean extraction error response.

Node 8 — Prepare Gemini Request

Code node. Retrieves jd_text from the Input Validation node and resume text from the extraction output. Validates both are present. Assembles the complete prompt string requesting a single JSON object containing all evaluation fields including the email draft.

Node 9 — Basic LLM Chain (Gemini)