Personal AI hub,
built from scratch
I design and operate a production-grade automation platform — ETL pipelines, LLM-powered agents, and a unified data layer — all running on infrastructure I own and maintain.
One platform, two purposes
yifeng.ai is my personal AI command center and a living portfolio. Ingestion workflows normalize email, SMS, and feeds into PostgreSQL; domain-specific agents read that data and produce actionable items — automatically.
Normalize first, decide later
Every signal flows through the same ETL pipeline before any agent touches it. Add a new agent without changing ingestion. Add a new source without changing agents.
Domain agents in production
Each agent filters the unified event store by domain tag, applies specialized logic, and outputs structured action items.
PM Agent · Property Management
Parses utility and vendor invoices from rental property emails, extracts amounts and dates via Gemini, records expenses, and surfaces payment deadlines.
domain: propertyJob Agent · Career Pipeline
Tracks interview invites, networking events, offers, and market signals. Maintains an application pipeline and drafts follow-up actions.
domain: jobExtensible by Design
Finance, calendar, and health agents plug into the same data layer. New domains require zero changes to the ingestion pipeline.
domain: *Built lean, hosted at home
The entire stack runs on a Synology NAS behind DDNS and a reverse proxy — no cloud vendor lock-in for core infrastructure.