Ahmad Maulana Malik Fattah

Data & AI Platform Engineer

I keep large analytical databases alive and make them answerable by dashboards, APIs, and agents.

Technical Lead at Bangunindo Teknusa Jaya for Indonesia's Ministry of Home Affairs: a national data platform for presidential-priority programs, and SIPD analytics for budget allocation and realization across local governments. Before that, a ClickHouse warehouse at Telkomsel ingesting 5+ TiB/day behind 300+ Apache NiFi pipelines.

Focus

The Platform

Click to filter.

Source SystemsOperational DatabasesETLBatch and StreamAnalytical StoreWarehouseDashboardReports and ChartsAPIsQuery and ServeAgentic AIAgents Over DataPlatform & OpsContainers, Config, and Observability

Stack and Proficiency

Grouped by where it sits in the platform. Proficiency: production / personal projects / exploring.

Source Systems

  • PostgreSQLproduction
  • MongoDBpersonal projects

ETL (batch & stream)

  • Apache NiFi — 300+ pipelinesproduction
  • Apache Kafkaproduction
  • Hadoop CDP / Hiveproduction
  • Data Scrapingproduction

Orchestration

  • Apache Airflowpersonal projects
  • Google Cloud Composerpersonal projects
  • dbtexploring
  • Mageexploring

Object Storage

  • Object Storage (MinIO, s3-compatibles)production
  • Google Cloud Storagepersonal projects

Analytical Store

  • ClickHouse — 5+ TiB/day warehouseproduction
  • BigQuerypersonal projects

Cache

  • Redisproduction

Dashboard

  • Apache Supersetproduction
  • Looker Studiopersonal projects
  • Streamlitpersonal projects

APIs

  • FastAPIpersonal projects

Agentic AI

  • LLMs (OpenAI / Anthropic / Gemini)production
  • FastMCPproduction
  • Keraspersonal projects
  • Transformerspersonal projects
  • Dify AIexploring
  • NVIDIA NIMexploring

Vector Store

  • ChromaDBpersonal projects

Platform & Ops

  • Linuxproduction
  • Dockerproduction
  • GitHub Actionspersonal projects
  • Ansibleexploring
  • Kubernetesexploring

Observability

  • Grafanaproduction
  • Weights & Biasespersonal projects

Languages

  • Pythonproduction
  • SQLproduction
  • Shellproduction
  • Goexploring

Projects

  • Telkomsel: Near-realtime Data Warehouse

    ETL · Analytical Store · closed source · Professional

    Data engineer on Telkomsel’s near-realtime warehouse: ClickHouse ingest at 5+ TiB/day behind 300+ Apache NiFi pipelines.

    Sources arrived as Kafka streams, SFTP drops, Hadoop CDP, and scraping jobs. NiFi owned ingestion, marts, and self-monitoring; ClickHouse was the analytical store the rest of the stack queried.

    The hard parts were operational: migrating to replicated engines without stranding consumers, recovering from node failures, repairing replication mismatches, and restoring from backup when the cluster lied about its own health.

  • Ministry of Home Affairs: Presidential-priority Programs

    ETL · Analytical Store · Platform & Ops · Dashboard · Agentic AI · closed source · Professional

    Technical lead on progress-monitoring dashboards for presidential-priority programs, including how the public sees those programs and Kemendagri itself.

    Dashboards track delivery of presidential-priority programs. The same serving layer supports monitoring of public perception on the programs and of the ministry.

    After Kemendagri licensed the platform, this team stays on for technical support: warehouse and BI uptime, and unblocking software and data engineers. An MCP server exposes Superset so agents can ask the same data the dashboards already serve.

  • Ministry of Home Affairs: Analytics for Local-government Budgets

    ETL · Dashboard · Analytical Store · closed source · Professional

    Technical lead on analytics over SIPD: budget allocation and realization for Indonesia's local governments. Organize team to brings the data from Postgres into ClickHouse, then dashboards.

    Operational systems at each local government remain the source of truth. This work is the analytics path: Postgres into ClickHouse, then dashboards that compare plan versus spend across regions.

    The job is correctness under messy source systems — reconciling warehouse figures with the originating databases, and keeping the serving layer honest when schemas and stages change.

  • BigQuery ML for Eco-Friendly Logistics

    Analytical Store · Dashboard · public · Medium (opens in a new tab)

    Assign each customer to the nearest warehouse in BigQuery, then forecast next-month stock with BigQuery ML and show it in Looker Studio. Assist business to have shorter hauls, lower logistics cost and lower carbon foot-prints.

    Built for a 1000StartUpDigital × Blibli workshop challenge: optimize which warehouse ships to which customer, then recommend stock levels for the next month. Distance ranking and BigQuery ML forecasting feed a Looker Studio dashboard stakeholders can read without opening SQL.

    Read on Medium

  • Superset Dev Tools

    Dashboard · APIs · Agentic AI · closed source · Professional

    No more wasting time on low-level maintenance. Effortlessly swap datasets and bulk-resave 100+ charts. For sure it's all reversible.

    When a warehouse schema moves, hundreds of Superset charts become wrong at once. Manual edits do not scale and leave no audit trail.

    Built tooling that swaps datasets and bulk-resaves charts with dry-run, apply, and revert; ensuring migration is a reviewed change instead of a weekend of clicking. The same surface later became an MCP server so an internal AI workspace can inspect and operate Superset without giving agents raw DB credentials.

  • Universa AI: Multi-Agent Framework

    Agentic AI · closed source · Professional

    Co-authored a Graph-of-Thought and MCP-like agent framework and managed three intern squads delivering NLP and RAG work.

    At Universa AI I co-authored the foundational agent base classes, the tool layer, and graph-based orchestration on a model-agnostic Graph-of-Thought and MCP-like framework.

    The goal was structured reasoning from non-reasoning LLMs: agents that produce functional software artifacts instead of vibes. Alongside that, I ran three intern squads (15+) shipping NLP classifiers, RAG pipelines, fine-tuning, and synthetic data orchestration.

  • Mood Tracker

    Agentic AI · public · GitHub

    Energy-less mood journaling with AI reflection. You can blatantly tell how you feel, it will help naming – and hopefully process – your emotions.

    TLDR: Exhaustive-depressive-all-of-the-above day suck up your energy. This app designed to help you keep track your mood swings easily, effortlessly, energy-less-ly.

    A Telegram bot for daily mood journaling that stores entries and assisted by AI for reflective prompts. Built end-to-end: bot UX, API, persistence, and model integration. A small system that has to stay reliable because people use it when they are not at a desk.

  • Emotion Classifier

    Agentic AI · public · GitHub

    Facial emotion classification web app. MTCNN face detection, VGG models, FastAPI + Streamlit.

    Face detection with MTCNN, emotion classification with VGG-family models, served behind FastAPI and a Streamlit UI. A full path from model to something a non-ML user can try in a browser.

  • HIMSIKA Event Feedback Analysis

    ETL · Analytical Store · Dashboard · public · GitHub

    End-to-end event feedback pipeline. Bringing unstandardized scattered data with Airflow to BigQuery, and landed as analytical dashboard on Looker Studio.

    Student-organization feedback collected, transformed, and published. Airflow orchestrates the load into BigQuery; Looker Studio surfaces the results so organizers can act without waiting on a one-off spreadsheet.

  • Chinook ETL

    ETL · public · GitHub

    ETL into a BigQuery warehouse modeled to answer business questions.

    Chinook sample data modeled into BigQuery so the warehouse answers business questions instead of mirroring the OLTP schema. It was my training ground for choosing data modeling.

  • Serverless ETL for Bloomberg Data

    ETL · public · GitHub

    Serverless ETL from Bloomberg API into BigQuery.

    Serverless path from Bloomberg market data into BigQuery. With scheduled extract, transform, and load without maintaining a long-lived worker.