Hong Phuc builds developer tools, automation systems, and workflows for the AI engineering stack, keeping quality visible.
$ ao delegate "debug failing test" --to Codex
lint: clean typecheck: ok tests: 142 passed build: 31s
$ rtk gain --history --scope personal-os
$ hermes cronjob run -- digest ships Mon/Wed/Fri
Open-source toolkits and systems built for real workflows. Each project carries its own design constraints, verification strategy, and operational scars.
Integration-first coordination protocol for 6 AI coding agents. Repo-native file claims, handoffs, routing, and delegate with auto-routing — ao CLI, MIT license, public template repo.
6
agents
4
milestones
MIT
license
Orchestrates Hermes, Codex, and Claude Code agents with shared skill libraries, RTK token compression, persistent memory, and systematic evaluation loops.
3
agents
100+
skills
60-90%
compress
Dockerized Flask dashboard for turning Threads and Reddit conversations into short-form videos. The public surface exposes the library and settings views while blocking mutation-heavy actions.
4000
port
Docker
image
demo
mode
Beyond code — these are running surfaces with users, schedules, and real operational footprints.
Hermes-native OSS intelligence pipeline: curl+jq fetcher, agent-written summaries, cron-scheduled Telegram delivery — zero API keys, zero external LLM cost, no browser automation.
120+
repos
Mon/Wed/Fri
cadence
$0/run
cost
Public demo for generating Vietnamese short-form news videos from URLs or prepared scripts. The demo runs in read-only mode so visitors can inspect the workflow without spending API credits.
9:16
format
Node
runtime
demo
mode
The center of gravity is still craft: simple boundaries, explicit tradeoffs, fast feedback, and tools that make good behavior easier.
01
I like interfaces that make hard work feel understandable: sharp boundaries, useful defaults, observable behavior, and escape hatches when the abstraction leaks.
02
The best AI workflows keep the engineer in the loop. Agents should compress toil, preserve intent, and leave a trail that future-you can trust.
03
Developer experience is not polish at the end. It is architecture, naming, ergonomics, feedback loops, and the discipline to delete what does not earn its place.
Markdown-backed essays ship with static metadata, RSS output, and focused reading pages.
Terminal setup, AI workflows, local LLM experiments, and quality gates are treated as one system.
Terminal, editor, shell — the surface I live in every day.
Agent runtimes and orchestration primitives for autonomous workflows.
Local inference, fine-tuning, embeddings, and retrieval pipelines.
Verification gates that run before every ship without thinking.
I like projects with hard constraints, clear taste, and enough technical depth to reward careful engineering.