01
Enterprise AI transformation
Start from high-value activities, surface and capture know-how, then rebuild it as Human + AI workflows. The point isn't just the model — it's closing the loop from assistance to delegation to execution.
Night voyage · unhurried
Solutions Architect / AI × Data
Hi, I'm Hobson, a solutions architect. Over fifteen years I've moved from Oracle DBA work, distributed databases and core financial systems to enterprise AI and agent engineering. My work has always sat where business, customers and technology meet: untangling complex problems, capturing the know-how people carry in their heads, and getting solutions running in real environments.
Turning years of data and architecture experience into delivery capabilities for the AI era — reusable and verifiable.
01
Start from high-value activities, surface and capture know-how, then rebuild it as Human + AI workflows. The point isn't just the model — it's closing the loop from assistance to delegation to execution.
02
Ongoing, hands-on work across context, tools, memory, MCP, RAG and evaluation, so agents have clear boundaries, reliable tools, observable runs and verification before anything ships.
03
From Oracle and PostgreSQL to distributed and cloud databases: migration, disaster recovery, performance and hybrid search — and how data infrastructure becomes a dependable foundation for AI.
Personal experiments, packaged as open-source projects that others can reuse, verify and maintain.
OPEN
A local-first macOS menu bar app that tracks and limits the time you actually spend in AI coding tools, with daily limits, a trends dashboard and shareable cards.
macOS / Python
OPEN
Direct and render short films in code: each film is a web page captured frame by frame in headless Chrome, with voice-over, music and subtitles on one timeline and ffmpeg doing the final cut. Comes with a browser-based studio.
Video / Node.js
OPEN
A per-pane conversation history rail for Herdr. It lists the turns in the current agent session and jumps Grok's scrollback to the prompt you pick. Each pane keeps its own history, and the rail closes with the pane.
Herdr / Plugin
OPEN
A local content workspace for solo creators that ties together AI-assisted drafting, assets, layouts, multi-platform publishing and Obsidian.
Next.js / AI
OPEN
A self-hosted assistant for fetal heart monitoring (CTG/NST): analysis, records and trend annotations, driven by a rules engine and deployable on Docker or a NAS.
Health Data / Python
OPEN
A small, offline macOS utility in Swift that nudges the mouse by one pixel after a visible countdown once you've been idle.
macOS / Swift
OPEN
A local web remote for Apple TV that runs on macOS, with device discovery, pairing and everyday playback controls.
macOS / Python
GitHub @yafeishi · 36 public repos · Oracle / PostgreSQL / TDSQL · synced 2026-10-10
From database operations and product work to data and enterprise AI solutions.
2022—
Database selection, PoCs, migration and disaster recovery for financial services and beyond — plus putting agents, knowledge bases and evaluation to work in real delivery.
2016—2021
Product planning, delivery standards, presales and domestic-stack replacement for a distributed database. Delivered internal training to 1,000+ attendees and spoke at PostgreSQL community events.
2011—2016
Deployment, architecture, tuning and training for CRM, BOSS and other core systems at provincial telecom carriers; later presales, PoCs and targeted tuning for a SQL review product.
On enterprise AI, engineering that ships, and how things actually get done.
The key to enterprise AI isn't plugging in a model. It's getting context, data, tools, permissions, governance and evaluation to work as one closed loop.
Architecture
Make tacit know-how explicit, then follow Know-how → Workflow → Agent → Capability to turn individual experience into organizational capability.
Transformation
Knowledge and context first. Clear tool contracts. Closed verification loops. Shipping over theorizing.
Engineering
Think, yes — but above all, act. Progress only comes from moving.
Weekly journal
2026
A per-pane conversation history rail for Herdr: it lists the turns in the current agent session and jumps straight back to the matching prompt in Grok.
Herdr / Plugin
2026
Years of Obsidian notes turned into a first-person agent that speaks in my voice: local hybrid search, long-term memory, a writing-style profile and personal context that keeps updating.
AI / Memory
2026
An evolving playbook on MCP, skills, harnesses, sandboxes, observability and layered evaluation — the engineering constraints an agent needs to go from demo to dependable delivery.
Agent / Eval
2022—
Selection, PoCs, migration and disaster recovery for securities, futures and insurance firms — turning distributed databases and domestic-stack replacements into systems that go live and stay maintainable.
Data / Database
Running
Dai Jiansong / Zheng Jiaxuan
Running
Herbert Steffny
Tech
Silvia Botros / Jeremy Tinley
Tech
Li Feifei et al.
Growth
Eric Jorgenson / Naval Ravikant
History
Yuval Noah Harari