Work

Built for production, not for demos

Distributed financial applications, large-scale data migration, geospatial APIs and platform infrastructure.

01 BRAC IT, Jul 2025 – present Python, Celery, PostgreSQL Client name withheld under NDA

Parallel Migration Pipeline

A distributed financial application, redesigned from sequential processing into a parallel, horizontally scalable pipeline.

Before

35K

group records in about 14 hours

After

3.26M+

records migrated in about 5.5 hours

The problem: Records moved through the pipeline one after another. 35,000 group records took nearly 14 hours, and recovering from a failure was manual.
The challenge: Add concurrency without losing or double-processing a record, and recover from failures without an operator watching over it.
What I built: Records are distributed across parallel workers and persisted to PostgreSQL, with automatic recovery of stuck jobs, database optimization and live monitoring. Load-tested at about half of server capacity.
The outcome: More than 5.7 million Celery task executions in production at a 99.7% success rate, across 79 branches, peaking at about 300 records a second.

Legacy state against the resulting migration capability, not a controlled benchmark.

Read the case study

02 BJIT, World Bank / JICA, 2024 – 2025 Django REST, GeoDjango, Celery, Redis, AWS

DRIVER

Backend for the Dhaka Metropolitan Police road incident platform. Thirty years of data, one migration, zero loss.

Before

MS-DOS

thirty years of records, 1990 to 2020

After

data loss on a geospatial, ML-enabled platform

The problem: Road incident records from 1990 to 2020 were locked inside an MS-DOS-era system that couldn’t be queried, analyzed or extended.
The challenge: Transform three decades of legacy records into a clean relational, geospatial model. The acceptance bar was zero data loss.
What I built: Automated the extraction, transformation and loading into a Django REST and GeoDjango platform on AWS, reconciling records against the source.
The outcome: The migration completed with 0% data loss, and machine learning models now forecast danger and red-zone areas. The work earned a performance bonus.

Read the case study

Also built

BJIT

Battery Management System

Lead backend engineer for a monitoring and automation platform. Refactored the codebase for API responsiveness, automated unit testing, and built CI/CD with AWS SAM and CodePipeline.

FastAPI, Lambda, DynamoDB, EventBridge, AWS CDK

BJIT

Startup Studio

One of Japan’s largest startup ecosystem platforms. Profiled login and signup from several seconds to under one, and cut cloud spend by about 40%.

Django, GCP, Redis, Celery, Cloud SQL

BJIT

Zenrin Map

High-performance geospatial map API. Caching and database optimizations to keep latency down under heavy traffic.

Django REST, GeoDjango, PostGIS, Redis

BJIT

BJIT ERP

Led and mentored the team that migrated more than 350,000 records across four modules from Odoo 11 to Odoo 16, inside a one-month deadline.

Odoo 16, Python, PostgreSQL, Docker

BJIT

NDBS DataHub

AWS Step Functions and Lambda moving data between RDS, CSV and S3, with automated file handling and integrity checks.

Step Functions, Lambda, RDS, S3

BJIT

MEMO2

Led the backend for a sound recognition system: audio fingerprinting with Dejavu and audfprint on scalable AWS infrastructure.

Django REST, EC2, Lambda, RDS, Docker