Turning manual processes
into production pipelines.
Data pipelines, scheduled reports, internal tools, and LLM agents that ask before they act. Built in Python, Postgres, Next.js, and the Claude API.
Open to full-time, contract and part-time project work, remote (UTC+8).
Selected works.
Async ETL Platform
Async multi-pipeline ETL into Postgres: about 14 scheduled pipelines across 25 GitHub Actions workflows

Workforce Report-Compliance Platform
Internal platform for daily-report approvals, reminders, management PDFs and extracts, with write-back to the vendor API as each approver
Quote Automation
Internal quoting tool that took accounting from 20-30 quotes a day to over 100: review queue, batch PDF-to-Drive and Gmail-draft emailing
Experiments, pilots and personal tools · 6
What I build with.
Data & Pipelines
AI & Orchestration
Where I’ve run things.
Hired as a Data Analyst; the work is data and AI automation engineering, and I build the platform solo. Replaced the team's manual pull-and-clean routine across API, Google/Microsoft Drive, Sheets and email with about 14 scheduled pipelines into a Supabase warehouse (12.2M+ rows across 111 tables), then automated the PDF reports and built dashboards on top. I also ship tools for other departments: a quoting app that took accounting from 20-30 quotes a day to over 100, and a workforce report-compliance platform used for daily-report approvals. On the AI side, an approval-gated agent that diagnoses failed pipeline runs is in production, and DARA (natural-language reporting) is a working prototype.
Demand planning for the HVACR business. Migrated a manual Excel forecast-tracking system into a real-time Power BI dashboard and wrote a Python script to clean, merge and consolidate the raw demand data, cutting analyst hours, improving accuracy, and giving stakeholders up-to-date insight instantly instead of waiting on hand-built reports.
QA and compliance in precision manufacturing, where preparing quality reports meant manually collecting and cleaning data from multiple sources before any analysis could begin. Automated that collection and cleaning with Excel Macros, VBA and Power Query (87.5% less reporting time), which freed the team for actual insight and fed Lean Six Sigma Kaizen work that cut defects by 4.9%. Earned the company's Top Management's Choice Award (2022) for the cumulative result: higher production rate, less time lost to reporting, and fewer rejects.
Worked the front line through my Industrial Engineering degree. Those were the years that taught pace, reliability, and showing up before the rest of the résumé existed.
Three things I optimize for.
Systems that operate themselves
I'd rather spend a week making a pipeline self-healing than answer the same 2 AM page twice. Auto-remediation, retries with backoff, and runbook-aware agents beat heroics.
Validate counts as a first-class signal
The worst bugs don't error. They silently return fewer rows. I treat row-count reconciliation and data-quality checks as core features, not afterthoughts.
Only count it learned once it ships
I keep a structured roadmap, but a course isn't done until the idea lands in a production system: evals, safety rails, and observability included.
Got manual work that should run itself?
Open to full-time, contract and part-time project work, remote (UTC+8). Email me what the process looks like today.