Building an Autonomous AI Job Hunter: A Step-by-Step Technical Guide

In this post, I’ll walk you through how I built an autonomous AI agent that scouts for jobs, analyzes them against my specific tech stack, and notifies me on Telegram when a high-quality match is found.

1. The Architecture

The system is built using the Hermes AI Agent framework. It acts as an orchestrator that connects a Large Language Model (the brain) to local files and external web tools.

Core Components:

• Brain: Google Gemini 1.5 Flash (via API).
• Storage: Local Markdown and LaTeX files for resume and skill definitions.
• Search Engine: Firecrawl / TinyFish API (to bypass bot detection on job boards).
• Communication: Telegram Bot API for real-time alerts.

---

2. Step-by-Step Implementation

Step 1: Preparing the Source of Truth

Traditional PDFs are hard for AI to parse reliably. I moved my resume from an Overleaf LaTeX project into a plain text/LaTeX file locally. This ensures the AI understands the semantic meaning of my projects, like my AI Book Agent and my experience at Earnest Data Analytics.

Step 2: Configuring the Agent’s "Hands"

To prevent being blocked by LinkedIn or Naukri, I configured a specialized web-scraping provider.

• Tool: Firecrawl API.
• Integration: Configured via ~/.hermes/.env to allow the agent to "see" the live web as structured data rather than a rendered UI.

Step 3: Defining the "JobScout" Skill

I wrote a custom "Skill" in Markdown that defines the agent's behavior. Instead of a simple search, it uses a Reasoning Loop:

1. Search: Query for "Node.js" and "Next.js" roles.
2. Analysis: Compare the job description against my specific backend experience.
3. Scoring: Only proceed if the match score is > 7/10.

Step 4: Automating the Workflow

Using the Hermes internal scheduler, I set the agent to run autonomously every morning. It wakes up, scans the web, performs its analysis, and shuts down, only notifying me if it finds something worth my time.

---

3. Challenges Overcome

• Bot Detection: Standard headless browsers were blocked instantly. Switching to an API-based search provider (Firecrawl) was the breakthrough.
• Context Management: Ensuring the AI didn't "hallucinate" skills meant providing the raw LaTeX source as context, which contains more technical detail than a standard PDF.

4. The Result

I now have a personal recruiter that works 24/7. It doesn't just find links; it explains why a job is a good fit based on my past work.

---

Built with: Node.js, Hermes CLI, Gemini 1.5, and Firecrawl.