01 · Capability
AI skill engineering
Adapt prompts and tool instructions to different model capabilities, then validate continuously against current pipeline data and team feedback.
AI Agent & Developer Tools Engineer
I build AI tools that turn engineering questions into validated results.
I design reusable AI skills, model-evaluation workflows, MCP services, and production ML systems. My focus is a faster experiment loop—without losing code quality, reliability, or engineering context.
Los Angeles, California · Open to opportunities in Taiwan
AI-native engineering systems
Compress the experiment loop
The highest-leverage AI tools connect problem decomposition, hypothesis formation, experimentation, evaluation, and feedback into one reliable loop. That system-level view guides how I build skills, workflows, and infrastructure for other engineers.
- 01Decompose
- 02Hypothesize
- 03Experiment
- 04Evaluate
- 05Feed back
02 · Workflow
Agentic development
Use Codex and Antigravity as AI IDEs, Ghostty as the terminal, and Obsidian traces and todos to preserve decisions, support regression, and prevent agent drift.
03 · Judgment
Engineering integration
Shift effort toward business goals, code review, pipeline performance, and integration fit—choosing the solution that improves the whole system.
Taboola · 2017–Present
AI tooling and production ML outcomes
LA Team Lead, Senior Algorithm Engineer, and Data Scientist building tools for engineers and ML systems for large-scale products.
Developer-facing AI automation
Reusable skills across the R&D lifecycle
- Problem framing: combine domain know-how with dataset and database-query patterns to decompose issues, test hypotheses, and recommend solution paths.
- End-to-end trace: follow event-model pixels through the full funnel, identify exclusion stage and reason, and verify training inclusion.
- Offline model evaluation: automate Notebook Hub setup and event/pageview comparison from simple dataset, date-range, and trained-model inputs.
Production ML business impact
Scale, quality, and measurable outcomes
- Predictive integration: PySpark and Java Spark ETL serving 400M daily users and 12M conversions; contributed to 60%+ conversion-rate improvement.
- Content moderation: reduced manual review by 80%; Mistral-7B improved accuracy by 5–10%.
- Quality and topic insight: BERT classification increased revenue 2–3%; multimodal safety reduced review time 10–20%; topic insights drove 10–15% traffic growth.
Founder · AI Product Engineer · 2026
DappGo
An AI-powered market intelligence product family spanning Taiwan equities, U.S. equities, and options.
Explore dappgo.comMCP platform
Built a TypeScript MCP server with seven AI-callable tools and compact, typed contracts so models can discover, query, and compare multiple data products through one interface.
Service scale-out
Designed model/server interoperability across stdio and Streamable HTTP with caching, stale fallback, concurrent-fetch deduplication, authentication, rate limiting, structured logs, tests, Docker, and Cloud Run.
Selected public work · Active in 2026
GitHub repositories
Public artifacts from current AI skills, developer tooling, and DappGo data-product work.
options-daily-report
Automated options strategy reports with Black-Scholes analysis, AI market commentary, and scheduled publishing.
us-stocks-daily-report
Published U.S. equity reports, schema-validated JSON, and dashboard data produced for the DappGo mobile app.
tw-stocks-daily-report
Published Taiwan equity reports and CDN-delivered structured data from a multi-source analysis pipeline.
dappgo-stocks-shared
Pure TypeScript types and tested utilities shared across the Taiwan and U.S. stocks product family.
video-subtitle-skill
Scene-aware AI workflow for contextual subtitles, animated text overlays, reusable configuration, and video rendering.
claude-code-usage-statusline
A compact Claude Code status line for model, context-window, Git, and rate-limit visibility inside the coding loop.
Technical range
From model workflow to production platform
AI & agents
MCP · reusable AI skills · prompt and tool engineering · tool calling · cross-model validation · LLM workflows · model evaluation
Programming
Python · TypeScript / JavaScript · Java · C / C++ · PySpark / Spark · React / React Native
Data & ML
Airflow · Kafka · Jupyter / Notebook Hub · BigQuery · HDFS / HBase · TensorFlow · scikit-learn · BERT · Mistral
Delivery
Docker · Kubernetes · Cloud Run · AWS · Jenkins · GitHub CI/CD · testing · code review · observability