Data platforms built end to end: ingestion, warehouse, orchestration, and the reports people open every day.

Independent data and analytics engineering practice, run by Idzharul Huda

I build new data platforms from scratch and fix the ones already running. At Rey.id I lead the batch analytics platform: 121 tables ingested hourly, 599 dbt models, and 908 automated data tests. RaschLab, my psychometric analysis platform, is the product side of the same work.

Primary delta: slot time reduction

96 %

Less BigQuery slot time on critical datamart model

BigQuery slot time reduction from 100 to 4 Slot time decreased from 100 baseline down to 4 residual across partitioning, clustering, and pruning steps. 100 01Baseline -52% 02Partition -28% 03Cluster -16% 04Prune 4 05Residual

Scope: one production datamart model, same query and data volume before and after
Source: query audit, 2024

Services

How engagements usually start.

  • New platform builds

    Greenfield data platform: source ingestion, warehouse modeling, orchestration, tests, and documentation, from the first source to the first report.

  • Rescuing a running platform

    Cost and scan reduction, model consolidation, test coverage, and the checks that catch a job which succeeds but writes nothing.

  • Delivery into existing tools

    Reports into spreadsheets, dashboards, scheduled files, and APIs, plus the internal layer where non-data teams file a data request, browse the asset catalogue, and see whether the pipelines behind it are healthy.

  • Assessment data analysis

    Rasch and IRT item analysis: difficulty, fit, dimensionality, and instrument-level reporting, delivered through RaschLab.

Products

One product for teams that run large-scale item analysis, and one public hub where every published figure is recomputed before it ships.

RaschLab

Rasch and IRT item analysis as a web platform: item difficulty and fit, dimensionality checks, option-level diagnostics, and instrument-level reports. It replaces legacy desktop psychometric workflows.

  • Item difficulty, fit and dimensionality on one logit scale
  • Instrument-level reports a reviewer can sign off on
  • Runs on CSV or Excel response data
WRIGHT MAP · PERSONS AND ITEMS ON ONE LOGIT SCALE051015-10+1+21234567misfit threshold 1.50 (item fit)personslogit
Figure: RaschLab output, test file of 60 x 7; item codes as the app prints them, hardest last: 1 I-007, 2 I-004, 3 I-005, 4 I-008, 5 I-006, 6 I-001, 7 I-002.

Public Data Dashboard

Data stories built from official public sources: an Indonesian earthquake catalogue taken from the USGS record, and PISA 2025 science results from the OECD annex tables. Each page is a static build, and every figure on it is recomputed from the source files by an acceptance gate before it publishes.

  • Append-only earthquake catalogue, rebuilt and re-verified every day
  • Static pages: one request, no runtime dependencies, no API keys
  • Every published figure recomputed from the source before deploy
M5+ EARTHQUAKE CATALOGUE · USGS · 1976–202610.517events recorded206 per year · median depth 35,0 kmbar length: log scaleM59.775M6662M775M84M9174,7% shallow (under 70 km) · 22,7% mid (70 to under 300 km) · 2,6% deep (300 km or more)
Figure: the M5+ catalogue as the hub publishes it, recomputed on every publish.

More builds, including incentive and operations reporting pipelines, are listed under Selected work below.

9x

faster processing on that same datamart model

Processing speed 1x to 9xIn the time the original query finished one run, the optimised query finishes 9 runs.Before1 runAfter9 runs123456789same time window, same data volume

Scope: full-refresh run, same data volume
Source: query audit, 2024

599 models

dbt models in production, with 908 automated data tests

599 dbt models by materialisation404 write a physical table, 190 are inlined, 5 build incrementally. One square per model.404 · 67%tables190 · 32%inlined5 · <1%incremental

Scope: 404 models write a physical table each run, 190 are inlined, 5 build incrementally; 233 models rebuild hourly and 124 daily
Source: platform measurement, Sep 2026

3.5 TB/day

scanned in the warehouse, fully on-demand

3.5 TB scanned per day against 2.7 TB storedThe warehouse stores 2.7 TB. Queries scan 3.5 TB every day, about 1.3 times the whole warehouse, roughly 105 TB a month.Stored in the warehouse2.7 TBScanned every day3.5 TB≈1.3× the whole warehouse, every dayOne month≈105 TBone block = one day

Scope: about 105 TB a month; the warehouse holds 2.7 TB and 1.5 bn rows across roughly 3,800 tables in 28 datasets
Source: platform measurement, Aug-Sep 2026

Selected work

Production pipelines, cost rewrites and automated delivery

Project / StackMeasured DeltaLink
01 Internal Data Request and Catalogue Platform
FastAPI, Postgres, Jinja
127 requests tracked end to end over three months, 49 of them in the busiest month, with 854 assets and 1,670 lineage edges kept current.Case study
02 BigQuery Pipeline Optimization
BigQuery, SQL, Partitioning
One datamart model: 9x faster, 96% less slot time, 58% less data shuffled.Case study
03 Dataverse: Conversational AI Analytics
LLM, Python, Streamlit
Upload a dataset, ask in plain language, get SQL and charts: no analyst in the loop.Case studySource
04 DBT + Metabase Self-Serve Analytics
dbt, BigQuery, Metabase
50+ reports moved to BigQuery, daily to hourly, with 40+ stakeholders self-serving.Case studySource
05 OCR + LLM Document Digitization
Python, YOLO, GPT-4
Thousands of hardcopy documents digitized, 70% efficiency gain, client evaluation demo shipped.Internal
06 Automated Data Quality Framework
dbt tests, BigQuery, Python
Automated checks and discrepancy detection: 50% less manual error resolution.Internal
07 Telehealth Cost-Effectiveness Study
Research, Statistical Analysis
Telehealth managed care at 7x the cost-effectiveness of conventional care for ARI.Case study

Automation modules and engines

Apps Script workbooks and the RaschLab analysis engine

Project / StackMeasured DeltaLink
08 Hourly Incentive Reporting Pipeline
Google Apps Script, Metabase, Zoho Analytics
Eight analytics feeds, hourly, into one workbook: every divergence audited.Case study
09 Incentive Performance Dashboard
Google Apps Script, JavaScript, Google Sheets
One URL that answers the current incentive week without spreadsheet wrangling.Case study
10 RaschLab engine
Python, Item Response Theory, Rasch model
The engine behind RaschLab, checked row by row against a legacy tool on real datasets.Case study
11 Clinic Incentives Dashboard
Google Apps Script, Google Sheets, Formula dashboard
Apps Script + Sheets dashboard with a Config tab and formula-driven incentive targets.Case study

Data quality and refresh matrix

Automated test coverage and hourly table syncs

Tech stack

Daily driver and production tools paired with usage scope

SQL all transformation, 599 dbt models in the warehouse
dbt the modeling layer, run from CI
BigQuery 2.7 TB stored, 3.5 TB scanned a day
Airflow 21 hourly pipelines, self-hosted on Kubernetes
Looker Studio primary BI surface for the business
Google Sheets analyst extracts and finance reporting
Metabase dedicated dataset and its own refresh job, secondary to Looker Studio
Dataplex catalogue and data quality
Cloud Storage file staging for ingestion and outbound delivery
GitHub Actions 6 workflows: PR checks, hourly, daily, on-demand, docs
Python third-party API pulls and data jobs
Google Apps Script clinic and incentive reporting dashboards
Streamlit Dataverse, a public build
Zoho Analytics hourly incentive sync into a reporting workbook

Experience

Career trajectory in analytics engineering and data platform infrastructure

Order: newest first (left to right)

Aug 2026 - Present

Analytics and Reporting Lead

Rey.id | Jakarta, Indonesia

Lead the analytics and reporting function and own the batch analytics platform behind client reporting

  • Run the platform end to end: 25 ingestion pipelines, 21 of them hourly, loading 121 tables (about 2,900 table loads a day) into BigQuery with a change-set merge and a 2-hour look-back window
  • Maintain 599 dbt models and 908 automated data tests: 233 models rebuild hourly, 124 daily, all transformation as SQL inside the warehouse (no Spark), run from git through GitHub Actions

Aug 2022 - Aug 2026

Data Analyst (Analytics Engineer)

Rey.id | Jakarta, Indonesia

Enforced personal-data controls in the platform itself: column-level policy tags on name, email, phone, address, ID and financial fields, 29 tagged models, and a Dataplex catalogue of 776 tables and 1,473 relationships that blocks personal data from readers without clearance

  • Boosted report accuracy by 13% (82% -> 95%) by automating CHISS workflows with dynamic master mapping
  • Migrated 50+ reports from Google Sheets to BigQuery: daily -> hourly generation, 15 analysts freed from 2+ hours a day of manual reporting

Apr 2024 - Jan 2025

AI Engineer

Olvo.ai | Hong Kong, China (Remote)

Digitized thousands of hardcopy documents using OCR, YOLO, and GPT-4 - 70% efficiency improvement

  • Automated claims verification and pricing audits - 60% less manual review workload
  • Designed an LLM-based embedding system to standardize hospital formularies across institutions

How I work

I'm a Data Infrastructure Engineer who builds both the pipelines and the delivery layer around them. I engineer data systems that deliver insights wherever people need them: dashboards, spreadsheets, email, APIs, and chat.

At Rey.id, I cut BigQuery slot time on the reporting datamart by 96% and made the same model run 9x faster, built automated data quality checks that cut manual error resolution by 50%, and pioneered a dbt + Metabase framework that enabled 40+ stakeholders to self-serve analytics.

Previously as an AI Engineer at Olvo.ai, I built OCR and LLM pipelines that digitized thousands of hardcopy documents into structured datasets.

My goal is to create scalable, trustworthy data systems that empower teams to move faster with confidence.

Location
Jakarta, Indonesia
Current role
Data Infrastructure Engineer
Focus
Data infrastructure: analytics engineering, pipeline cost optimization, and delivery into the tools people already use
Timezone
Asia/Jakarta (UTC+7)

Work with me

For a new platform build, a rescue of an existing pipeline, or a question about any of the numbers above, email is fastest.