Project archive
Each project links to the GitHub repository and a short case page with delivery context.
Stop copying invoices and statements by hand from portals that don't have an API.
A Python scraping flow that logs in, reconciles what it finds, and turns finance's manual copy-paste routine into clean, BigQuery-ready records.
Finance stops fighting inconsistent ERP exports, even with more than one system in the mix.
dbt layers turn whatever each ERP exports into one canonical model, with reconciliation and incremental loads that survive a messy backfill.
Analytics on GCP that your team can actually maintain, not a platform that outgrows you.
Raw, stage, and analytics layers on GCS and BigQuery, built with cost controls so a spike in usage doesn't turn into a surprise bill.
A lakehouse on AWS that holds up when an auditor actually asks how the data got there.
Bronze, silver, and gold layers on S3, with IAM/KMS governance built in from day one instead of bolted on after a security review flags it.
Orchestration a second engineer can read, not a cron job only one person understands.
Asset boundaries and dbt transforms replace cron-script guesswork, with replay-safe runs so a failed job doesn't mean redoing everything by hand.
Find out which query is inflating your BigQuery bill before finance asks.
Turns BigQuery job telemetry into a ranked backlog, so tuning work has a clear next target and an expected savings number attached.
Paid media numbers that hold up hour over hour, without someone patching a spreadsheet.
Campaign, spend, click, lead, and conversion data synced idempotently, so a duplicate row doesn't quietly inflate the numbers marketing reports on.
Mobile attribution you can act on, not a feed everyone double-checks before trusting.
Installs, in-app events, and campaign costs normalized through a path built for mobile's specific quirks, so growth stops reconciling numbers by hand.
A GA4 export your team can build channel reporting on, instead of a raw dump nobody quite trusts.
Sessions, conversions, and traffic-source data normalized into stable models, with logs clean enough to survive a finance or growth review.
One ranked view of what's rising, instead of five open tabs and a gut feeling.
News, social, and search feeds pulled into one scoring layer, so a rising topic shows up with a number attached instead of someone's hunch.
An AI automation that survives contact with routing, retries, and someone asking why it did that.
Inbox ingestion, CRM task logic, and agent action records turn every automated decision into something with a paper trail, so you can pull up exactly why the agent did what it did.
