Job Hunter Agent

Summary

Role: Sole Developer

Team size: 1

Year: 2026

Time span: few hours

Tools:

TypeScript

GitHub Actions

Node.js

Google Gemini API

Upstash Redis

Discord Webhooks

Tags:

AI & AgenticRapid Prototyping

A lightweight, zero-maintenance autonomous pipeline that aggregates remote job postings, evaluates them against custom candidate profiles using Gemini, and pushes matches to Discord.

Overview

Built as a quick, practical automation project to streamline the remote job hunt across Europe. The agent runs autonomously every morning via GitHub Actions, eliminating the friction of manual job board surfing. It aggregates listings from multiple remote job APIs, pre-filters them against targeted engineering personas, evaluates candidate-job fit using Google Gemini’s structured JSON mode, and delivers scored summaries directly to Discord.

Core Features

Multi-Source Aggregation & Sanitization: Pulls daily listings across multiple public remote endpoints (Jobicy, Remotive, Arbeitnow), deduplicates cross-posted roles, and strips HTML markup and non-English postings via heuristic language filtering.
Persona-Driven Pre-Filtering: Supports multiple distinct career personas (e.g., Simulation Engineering vs. AI/Agentic Systems). Matches candidates using both title and full-description regex parsing to catch relevant roles with generic titles (e.g., “Senior Software Engineer” working on simulation physics or agentic workflows).
Structured LLM Match Scoring: Feeds candidate profile Markdown and job descriptions into Gemini using strict JSON Schema enforcement. The model outputs a 0–100 compatibility score, qualification matches, potential skill gaps, and an executive verdict.
Adaptive & Self-Healing Model Fallback: Built-in resilience handles rate limits, service disruptions, and upstream model deprecations. The system dynamically respects API cooldown timers, applies retry backoffs, and cascades across a prioritized pool of lightweight models. It automatically self-heals at runtime by intercepting upgrade suggestions from API responses, blacklisting retired versions, promoting winning endpoints, and discovering live alternatives directly from the model registry if defaults are exhausted.
Zero-Cost Scheduled Execution: Runs fully headless on GitHub Actions runners with no cloud hosting bills.
Serverless Deduplication: Persists seen job IDs in Upstash Redis using HTTP-based batch pipelining and rolling TTL expiration, guaranteeing postings are never evaluated or notified twice.
Real-Time Discord Alerts: Pushes color-coded Discord embeds for any role meeting or exceeding the match threshold (score ≥ 75), complete with direct links, score badges, pros, and gap analysis.