Skip to content

json_resume_embedding_approach.md

Latest commit

 

History

History
151 lines (119 loc) · 4.24 KB

json_resume_embedding_approach.md

File metadata and controls

151 lines (119 loc) · 4.24 KB

JSON Resume Embedding Approach

This guide explains how to enhance your Cloudflare Worker with a JSON-based resume embedding approach.
It avoids Vectorize neuron limits and lets you provide structured context to an OpenAI LLM.

Step 1 — Convert Resume to JSON

You’ll first need to convert your resume (e.g., Word .docx) into structured JSON. This will let you provide clean, sectioned data to the LLM.

Extracting from Word

  1. Save your resume as a .docx file.
  2. Parse it into JSON using Python:
from docx import Document
import json

doc = Document("resume.docx")

experience, skills, education, projects = [], [], [], []
current_section = None

for para in doc.paragraphs:
    text = para.text.strip()
    if not text:
        continue
    if text.lower().startswith("experience"):
        current_section = experience
        continue
    elif text.lower().startswith("skills"):
        current_section = skills
        continue
    elif text.lower().startswith("education"):
        current_section = education
        continue
    elif text.lower().startswith("projects"):
        current_section = projects
        continue
    if current_section is not None:
        current_section.append(text)

resume_json = {
    "experience": experience,
    "skills": skills,
    "education": education,
    "projects": projects
}

with open("resume.json", "w") as f:
    json.dump(resume_json, f, indent=2)
  1. Verify the output by opening resume.json. Each section should be an array of strings. Adjust the section headers in the script if your resume uses different labels (e.g., Work History instead of Experience).

Example JSON Output

{
  "experience": [
    "IT Lead at AAC Construction (2020–Present) – Managed Azure API Management, containerized workloads, and automation with Terraform.",
    "Microsoft Certified Trainer (2015–2020) – Specialized in Azure AI, DevOps, and cloud architecture."
  ],
  "skills": [
    "Azure, API Management, Terraform, Python, Cloudflare Workers, GitHub Actions"
  ],
  "education": [
    "B.S. in Computer Science, Northern Kentucky University"
  ],
  "projects": [
    "Developed AI-driven chatbot for client onboarding",
    "Built Hugo + Cloudflare Pages static site with OpenAI API integration"
  ]
}

Step 2 — Upload Resume JSON to Cloudflare KV

Create a KV namespace:

npx wrangler kv namespace create RESUME

Bind it in wrangler.toml:

[[kv_namespaces]]
binding = "RESUME"
id = "your-kv-id"

Upload JSON:

npx wrangler kv:key put --binding=RESUME "resume" "$(cat resume.json)"

Step 3 — Worker Logic

Fetch JSON from KV, flatten it into text, and pass it into the system prompt.

export default {
  async fetch(req, env) {
    const url = new URL(req.url);
    const userQ = url.searchParams.get("q") || "Tell me about Mark’s background";

    // 1. Get resume JSON
    const resumeJson = await env.RESUME.get("resume", { type: "json" });

    // 2. Flatten resume into text
    const resumeText = Object.entries(resumeJson)
      .map(([section, items]) => `${section.toUpperCase()}:\n${items.join("\n")}`)
      .join("\n\n");

    // 3. Build HAL system prompt
    const systemPrompt = `
You are HAL 9000. Answer calmly and concisely.
You may use the following resume data when relevant:

${resumeText}
    `;

    // 4. Send to OpenAI
    const res = await fetch("https://api.openai.com/v1/chat/completions", {
      method: "POST",
      headers: {
        "Authorization": `Bearer ${env.OPENAI_API_KEY}`,
        "Content-Type": "application/json"
      },
      body: JSON.stringify({
        model: "gpt-4o-mini",
        messages: [
          { role: "system", content: systemPrompt },
          { role: "user", content: userQ }
        ]
      })
    });

    const data = await res.json();
    return new Response(data.choices?.[0]?.message?.content || "I'm sorry, I can't answer that.");
  }
}

Benefits

  • No neuron limits (KV storage is free and generous).
  • Structured resume = easy to update without retraining.
  • Works well for a HAL-style assistant that remembers your background.