I spent a weekend building a personal AI agent that searches for jobs, drafts emails, and keeps me informed via WhatsApp. Here's what I learned.
The Setup
I used n8n (self-hosted workflow automation) running on a mini PC. The stack:
- n8n for orchestration
- Ollama for local LLM inference
- PostgreSQL for persistence
- Cloudflare Tunnel for external access
- Twilio for WhatsApp integration
What Worked
The agent can search multiple job APIs, aggregate results, and let me interact conversationally. "Find me senior AI roles in Europe" actually works.
Email drafting with approval flow is nice. The agent drafts, I review in chat, say "send it" and it goes.
What Didn't
Mass job applications are low ROI. The real value is in targeted outreach—finding companies I actually want to work for, finding the right people, and crafting personalized messages.
Also, local LLMs (7B params) struggle with complex tool calling. Cloud APIs (Gemini, Groq) are more reliable for agentic workflows.
Takeaway
Building is learning. Even if this agent never lands me a job, I understand agentic architectures better now. That's the point.