🌐

Data Science + Economics

Hello world, my name is Howard.

or 胡昊阳

Trying to build the future before it becomes obvious.

SQLPythonPower BITableauA/B TestingHypothesis TestingTensorFlowPyTorchGoogle CloudAWSOpenClawAzure SQLPythonPower BITableauA/B TestingHypothesis TestingTensorFlowPyTorchGoogle CloudAWSOpenClawAzure
Howard coding pixel avatar

Systems / Interfaces / Automation

Here are some of the fun stuff and side quests I’ve built or am currently building.

Flagship System

JARVIS

A personal AI operating system

Stark Industries was fictional. The idea wasn’t.

JARVIS began as a simple question: what if I had a real digital teammate I could talk to all day, not just another chat window?

Instead of typing prompts, I wanted to say things like: “Jarvis, what’s on my agenda today?” or “Research this topic and build a full execution plan.”

JARVIS is my attempt to build that system.

It combines voice interaction with a modular agent architecture capable of research, synthesis, monitoring, and action. Under the hood, specialized agents coordinate tasks like web research, information extraction, summarization, and automated execution.

Rather than a single chatbot, JARVIS functions as a personal AI operating layer, a system that can run workflows end-to-end, from gathering information to producing usable outputs.

The long-term vision is a persistent AI partner that understands context, manages complex tasks, and acts as an always-available digital teammate.

This prototype explores that architecture through OpenClaw-style autonomous browsing, LLM orchestration, and real-time voice interaction.

PYTHON · LLM APIs · AGENT ORCHESTRATION · WEB AUTOMATION · VOICE INTERFACES · OPENCLAW-STYLE SYSTEMS

AGENT AUTOMATION

Stevie Tech Broadcast

where headlines become yaplines

An AI news pipeline that researches, scripts, and edits so Stevie can clock in on time.

This started as an experiment in one question: how much of a news studio can I automate before it turns into slop?

The pipeline tracks story candidates, analyzes what actually matters, gathers screenshots and support visuals, writes the script, and builds the short around Stevie as the anchor.

The goal is not random AI noise. It is fast, watchable, context-rich news content with better media selection, cleaner pacing, and less manual babysitting.

PYTHON • OPENAI API • FFMPEG • PLATFORM APIS

Data Visualization

Analytic Dashboard Maker

From raw tables to decision-ready dashboards

Because manually rebuilding dashboards every week is character development I did not need.

This project came from seeing how much analyst time gets burned cleaning data and redoing the same visuals. I built a pipeline that ingests messy business data, extracts key metrics, and prepares dashboard-ready structures with much less manual work.

I focused on making outputs practical: clear KPI definitions, reusable transformations, and automated handoff into BI tools so teams can spend more time deciding and less time wrangling.

Python · Power BI · SQL · Workflow automation

Research Infrastructure

Curiosity X

Continuous monitoring of the internet

My personal intel feed, minus the 20-tab chaos.

I built this because I was spending too much time context switching between feeds, forums, and tech news. The agent continuously monitors sources and compiles concise briefings.

It prioritizes what matters, summarizes developments with clear action points, and pushes updates through messaging so I can react fast without manual hunting.

Python · RSS ingestion · LLM summarization

Howard Hu pixel avatar

Hi there

Talk to me about data, AI, analytics, and building useful systems!

Work

Education

Howard camera pixel avatar

Capturing moments, documenting stories, and experimenting with light. Pick a category and step into the full set.

Book a session with me @howiesflicks

Photo Set

Film

Analog frames, softer grain, slower timing.