Brian Cusuanto

I get handed the pipeline nobody wants to touch.

Trusted with architecture decisions and cross-team delivery. I take data from raw source to analytics-ready, batch and streaming, modelled properly, with CI/CD wired in so a change ships the same day someone asks for it.

  • 124M+records / day
  • 75%infra cost cut
  • 99.8%pipeline reliability
  • 35Ktxn / minute streaming

01 About

Data Engineer with 5+ years in banking and property tech, owning architecture decisions and leading delivery across teams. In practice the work is the jobs that fail silently, the lineage nobody can draw anymore, the bill that quietly tripled.

In property tech that meant a valuation pipeline that was quietly dropping 19 out of every 20 rows. In banking it meant 17+ sources folded into one datamart, a two-hour lead ETL cut to five minutes, and fuzzy matching over 2M+ records at 30% better accuracy.

Most pipelines break quietly. I build the ones that don’t, and most of my work is measured in what it removed: cost, latency, manual steps, and the phone call at 3 a.m.

02 Experience

  1. Sep 2025 — Present

    Data Engineer — Big Data, Streaming & Analytics

    Bank Negara Indonesia

    • Shipped a production Kafka + Spark Streaming pipeline handling 25–35K transactions/minute, with schema registry enforcing schema evolution and data contracts.
    • Delivered a unified datamart for 20+ business and campaign reports, cutting query turnaround from a full day to under 5 minutes over 134M+ daily records.
    • Built HC data architecture consolidating 17+ sources into a 5-layer medallion datamart, 10+ entity tables for AI-ready analytics on ~36K employees.
    • Held production Spark workflows at 99.8% reliability while improving job performance up to 92%.

    Kafka · Spark Streaming · PySpark · Hive/Impala · Schema Registry

  2. May 2023 — Nov 2025

    Data Engineer — Lead Automation & Campaign Analytics

    Bank Negara Indonesia

    • Automated the sales-lead ETL end to end (passworded Excel, FTP delivery, email alerts), taking processing from 2 hours to under 5 minutes.
    • Optimized fuzzy matching on 2M+ records with RapidFuzz + multiprocessing: accuracy +30%, runtime −50%.
    • Built a self-healing SQL retry layer — failure resolution time −80%, 100% query success without manual watching.
    • Owned delivery and mentored junior and vendor engineers to shorten onboarding.

    Python · PySpark · RapidFuzz · SQL · Bash

  3. May 2021 — Apr 2023

    Data Engineer — Data Marts & Query Optimization

    Bank Negara Indonesia

    • Automated daily/weekly/monthly datamart refreshes via CDSW — 100% on-time delivery, zero manual runs.
    • Partitioned Hive/Impala tables on Parquet: 3x faster queries on less storage.
    • Tuned slow SQL across Hadoop, cutting execution time up to 70% and serving 10+ business domains.

    Hive · Impala · Parquet · PySpark · Cloudera

Before data: civil engineering & construction — quantity surveyor, structural/finishing inspector, drafter (2016–2020) · STEM teacher, part-time (2013–2014).

03 Projects

Orchestration Migration: Composer → Windmill

Consolidated scattered Airflow DAGs into one Windmill execution model, cutting orchestration infra cost 75% (US$1,200 → US$300/mo).

Windmill · dbt · BigQuery · Python · GCP

Property Valuation Pipeline Rescue

Diagnosed and rebuilt a broken home-valuation ingest — completeness from ~2 to 38 rows/day (~19x), lifting downstream valuation accuracy.

Python · curl_cffi (anti-bot) · BigQuery · Windmill

Medallion Warehouse on BigQuery

Re-architected fragmented sources into a dbt-modeled 5-dataset medallion warehouse; worst-case root-cause debugging from 8 hours to under 15 minutes.

dbt · BigQuery · Medallion architecture

Lead-Matching Lineage Rebuild

Untangled a MATCH lead-matching lineage in 4 days: 113 → 69 stages, downstream consumers from a dozen+ down to 4.

dbt · BigQuery · SQL

Banking Loan Risk ELT

How do banks screen millions of loan records daily? End-to-end pipeline from raw files to risk-ready tables.

Airflow · PySpark · dbt · BigQuery · Docker

E-Commerce ELT on Cloud Composer

Production-shaped Airflow 2.x pipeline with CI/CD and Slack alerting, runnable end to end.

Cloud Composer · dbt · BigQuery · GitHub Actions

Three-Source Sales ELT

Extracts from API, Postgres and Cloud Storage into one modeled sales summary.

Airflow · Docker · dbt · GCP

Banking Streaming OLTP → OLAP

Real-time transaction stream landed into analytics storage without breaking schema contracts.

Kafka · Spark Streaming · Schema Registry

Massive Lead Assignment

Distributes 100K–1M leads across sales reps by geography and customer criteria, fairly, round-robin.

Pandas · PySpark · Hive · CML

High-Speed Fuzzy Matching

Millions of messy name records matched in parallel instead of overnight.

RapidFuzz · Multiprocessing · Python

All repositories on GitHub

04 Skills

Languages

Python · SQL · Bash

Orchestration

Airflow · Windmill · dbt · Dagster · Cloud Composer · Cron

Streaming & Big Data

Kafka · Spark Streaming · PySpark · Hadoop · Hive · Impala · HDFS · Databricks

Cloud & Warehouse

BigQuery · Cloud Storage · Cloud Run · GKE · Vertex AI · Snowflake · Medallion architecture · Parquet · Partitioning

DevOps & CI/CD

Docker · Kubernetes · GitHub Actions · Bitbucket Pipelines · Linux · Git

Databases

PostgreSQL · MySQL · Oracle Exadata · Teradata · HiveQL · ImpalaQL

Scraping & Automation

curl_cffi (anti-bot) · Scrapy · BeautifulSoup · Playwright · Selenium · RapidFuzz

Education — Data Science & Machine Learning, Purwadhika Digital Technology School (2020) · B.Eng. Civil Engineering, Pembangunan Jaya University (2013–2017)

05 Contact

If you’ve got a pipeline nobody wants to touch, I’d like to hear about it. Open to full-time or contract work, remote, hybrid, or on-site.