OLJ-Worker: Engineering an Autonomous Job Intelligence & Gemini AI Proposal Engine on Cloudflare Workers

Executive Summary

For specialized freelance architects and technical consultants, timing and proposal quality are decisive. Platforms like OnlineJobs.ph publish hundreds of postings daily across WordPress, front-end engineering, and full-stack development. However, manual job hunting presents significant friction: hours spent refreshing search feeds, strict daily application point limits that punish applying to mismatched roles, and hidden employer screening traps designed to filter out bot spam.

To solve this, we engineered OLJ-Worker: a serverless, autonomous edge application deployed on Cloudflare Workers. Running on a 30-minute cron schedule, the worker searches targeted keywords, persists authenticated sessions in Cloudflare KV, parses detailed job listings with Cheerio, evaluates fit (1–10 scale) using Google Gemini 2.5 Flash against a candidate knowledge base, generates concise human-sounding proposals, and alerts the engineer instantly via Discord and Telegram.

The Challenge: Navigating Platform Constraints & Bot Traps

Automating interactions with legacy job boards introduces critical architectural challenges:

System Performance & Operational Benchmarks

Comparing the manual candidate workflow against the autonomous Cloudflare Worker pipeline:

The Solution Architecture

OLJ-Worker was designed around modular services in TypeScript, adhering to single-responsibility principles and edge resilience:

1. Authenticated KV Session Manager

Rather than logging in during each run, the AuthService authenticates via HTTPS form payload, extracts session cookies, and stores them in Cloudflare KV with a 24-hour TTL. Subsequent requests reuse the cached session headers, preventing rate limits and login thrashing.

2. Cheerio DOM Parser & Screening Question Extractor

The JobParserService navigates to individual job postings and extracts structured JSON containing:

• Job title, company name, and listed budget/salary.
• Full sanitized job description and tech stack tags.
• Mandatory screening questions and custom application fields.

3. Gemini 2.5 Multi-Dimensional Fit Evaluator

The raw job data is passed to Google Gemini 2.5 Flash with structured system instructions referencing chad-sia-master-resume.md and chad-sia-tone-of-voice.md. Gemini returns a strictly validated JSON response:

// Structured Fit Evaluation Output Interface
export interface GeminiJobFitResult {
  fitScore: number;          // 1 to 10 scale
  fitReasoning: string;      // Concise rationale for score
  isSpamOrLowQuality: boolean;
  screeningAnswers: Array<{
    question: string;
    answer: string;
  }>;
  tailoredProposal: string;  // 150-220 word bespoke cover letter
}

4. Safety Guardrails: 60-Day Deduplication & Dry-Run Sandbox

To guarantee safety, the system implements:

• 60-Day KV Deduplication: Applied job IDs are hashed in KV namespace OLJ_KV with a 5,184,000-second expiration.
• Daily Application Cap: Limits live submissions to a safe threshold (e.g., 10 applications/day) to conserve points.
• Dry-Run Mode: When DRY_RUN=true, the worker executes full scraping, Gemini fit analysis, and notification dispatch while skipping final HTTP form submission.

5. Multi-Channel Discord & Telegram Telemetry

Whenever a job with a Fit Score ≥ 7 is detected, rich telemetry is dispatched immediately:

• Discord: Color-coded rich embed (Green for 9–10, Blue for 7–8) with direct links, budget summary, screening answers, and full draft proposal.
• Telegram: Instant Markdown alert delivered to mobile with single-tap link access.

Core TypeScript Implementation

Below is an excerpt from the Gemini scoring and proposal generation engine:

/**
 * Google Gemini 2.5 Flash Fit & Proposal Orchestration
 */
export async function evaluateJobWithGemini(
  job: JobDetails,
  profile: CandidateProfile,
  apiKey: string
): Promise<GeminiJobFitResult> {
  const prompt = `
You are evaluating a job posting for ${profile.name} (${profile.title}).
Core Expertise: ${profile.coreSkills.join(', ')}
Rate Target: $${profile.targetHourlyRate}/hr ($${profile.targetMonthlyRate}/mo)

Job Title: ${job.title}
Employer: ${job.employer}
Salary: ${job.salary}
Job Description:
${job.description}

Screening Questions:
${job.screeningQuestions.join('\n')}

Instructions:
1. Score fit from 1 to 10 (deduct points for non-WordPress/unsupported stacks).
2. Answer all screening questions accurately based on profile experience.
3. Write a concise, natural, human cover letter (150-220 words). No AI buzzwords.
4. Reference relevant live portfolio examples: ${profile.featuredPortfolioUrls.join(', ')}.

Respond ONLY with valid JSON conforming to GeminiJobFitResult schema.`;

  const response = await fetch(`https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash:generateContent?key=${apiKey}`, {
    method: 'POST',
    headers: { 'Content-Type': 'application/json' },
    body: JSON.stringify({
      contents: [{ parts: [{ text: prompt }] }],
      generationConfig: { responseMimeType: 'application/json', temperature: 0.2 }
    })
  });

  const data = await response.json();
  return JSON.parse(data.candidates[0].content.parts[0].text);
}

Production Impact & Results

Deploying OLJ-Worker on Cloudflare Workers transformed client acquisition into an automated, high-precision operation: