Abdullah Mazhar · data infrastructure
SELECT portfolio, count(*) AS professional FROM Blockchain WHERE day = today()
Data infrastructure · Karachi
I design, deploy and operate the infrastructure that makes very large, very messy datasets fast to query. Postgres and ClickHouse, across multiple chains.
| ch-01 | 12.1% |
| ch-02 | 12.7% |
| ch-03 | 9.0% |
| ch-04 | 13.9% |
| pg-01 | 13.7% |
| pg-02 | 12.7% |
| ix-01 | 12.7% |
| ix-02 | 13.2% |
Hover a shard for its range and replicas. Click to kill it.
Architecture
This is the shape of a deployment I actually run: edge, two app containers, a pooled Postgres pair, a replicated ClickHouse pair, and an indexer feeding it. Traffic is routed rather than animated — every packet picks a live node at each hop. Hover a component for its role and blast radius. Click to kill it. Take out pg-primary and the standby promotes. Take out caddy and everything fails, because it is a single point of failure and the diagram says so.
hover for detail · click to kill
cluster
Hover a component.
If it dies
Traffic reroutes around anything that isn't a single point of failure. One thing here is.
Work
Every entry below leads with the technical decision rather than a feature list, because the decision is the part that transfers.
Ingest service that decodes blocks and events from several networks into one columnar schema.
Idempotent batch insert keyed on (chain_id, block_number, log_index), so a replay after a reorg converges instead of double-counting.
Benchmark rig that replays a real query workload against candidate partition keys.
Turns schema design from an argument into a measurement. Same 200 queries, five partitioning schemes, one table of p50/p95/p99.
Watchdog that detects chain reorganisations and rewinds affected partitions.
Keeps a rolling window of block hashes and compares against head. When they diverge it drops and rebuilds only the affected partitions.
Streaming replication, WAL archiving and automated failover without a managed provider.
Patroni plus etcd on three small boxes. The interesting part is the write-gap during promotion, which is roughly 15 seconds and worth being honest about.
Layered transformation models turning raw block and log tables into query-ready marts.
Incremental models keyed on partition so a backfill touches one day, not the whole history.
The pipeline this site ships on: build, push, health-gate, cut over.
Health-gated cutover — the new container has to answer before the old one stops. Cheap blue/green without a control plane.
These six are placeholders, named after realistic infrastructure work so the layout is designed against plausible content. Swap them in lib/content.ts — the section is pure data.
Colophon
Sep 2025 — presentcurrent
Own the indexing and warehousing layer: on-chain data out of nodes and RPC endpoints, into ClickHouse and Postgres schemas built for analytical queries. Schema and partitioning work, ETL automation, indexer reliability, replay of missing blocks.
ClickHousePostgresSQLETLIBCEVM
Jul 2025 — presentcurrent
Enterprise ERP on Oracle APEX against Oracle 21c/23ai. Database structures and ERDs, plus LLM modules integrated into existing corporate systems and blockchain-enabled tables for immutable audit trails.
Oracle 23aiAPEXPL/SQLERD
Jul 2025 — presentcurrent
AI agents, conversational systems and automated workflows, deployed to AWS on Ubuntu with Docker.
PythonLLMAWSDocker
Jul 2024 — Aug 2024
Smart contract development and auditing with Foundry, contract-to-frontend integration on Ethereum.
SolidityFoundryNext.jsEthereum
Education
BSc Computer Science
Mohammad Ali Jinnah University
2023 — 2027 · Computer architecture, artificial intelligence
Pre-Engineering
Chiniot Islamia College
2019 — 2021
Certifications · 17
Contact
I'm interested in infrastructure work — indexing, warehousing, deployment, and the unglamorous parts of keeping data systems correct. Email is the fastest way to reach me.