Backend Engineering & Systems Design

Building systems that reduce operational friction

I design and build backend systems that automate manual workflows, eliminate integration chaos, and scale with business demand. Not just code — engineered solutions with measurable impact.

Start a discussion View selected work

What I Solve

Problems, not features

Every system I build starts with a specific operational problem. Here are the patterns I encounter most frequently — and how I address them.

Problem

Manual workflows

Teams spending hours on repetitive data entry, report generation, and cross-system updates that should run automatically.

Approach

Map the full workflow end-to-end. Identify automation triggers. Build pipeline with error handling, retry logic, and audit trails.

Outcome

85% reduction in manual processing time

Problem

Complex approval flows

Hardcoded approval chains that break when organizational structure changes. No visibility into bottlenecks or escalations.

Approach

Design a dynamic rule engine. Separate flow definition from execution. Add delegation, timeout escalation, and real-time status tracking.

Outcome

Approval cycle time reduced from days to hours

Problem

API integration chaos

Dozens of point-to-point integrations with inconsistent error handling, no observability, and fragile coupling between services.

Approach

Build an integration layer with standardized adapters, request/response transformation, circuit breakers, and centralized logging.

Outcome

New integration onboarding: weeks to days

Selected Work

Systems in production

Each project below represents a real operational problem, a specific approach, and a quantified outcome — not a feature list.

Internal Platform 2024

Dynamic Approval Engine

A multinational company's procurement approvals were hardcoded across 12 microservices. Any organizational change required code deployments.

Solution

Built a rule-based approval engine with a visual flow builder. Approvals are now data-driven configurations, not code. Supports delegation, parallel paths, and timeout-based escalation.

Stack

Node.js, PostgreSQL, Redis, RabbitMQ

Impact

Approval reconfiguration: 2-week dev cycle to 15-minute config change. Processing 2,400+ approvals/month.

Integration System 2024

Structured Integration Gateway

A logistics company operated 23 point-to-point API integrations with no standardized error handling. Failures were silent, debugging took hours.

Solution

Designed a centralized integration gateway with adapter pattern for each external service. Added circuit breakers, retry policies, request/response logging, and a monitoring dashboard.

Stack

Go, gRPC, PostgreSQL, Grafana

Impact

Mean time to detect failures: 4 hours to 2 minutes. New integrations onboarded in days instead of weeks.

Workflow Automation 2023

Financial Reconciliation Pipeline

A fintech startup's finance team manually reconciled transaction data across 4 payment gateways and 2 banking systems every business day.

Solution

Built an automated reconciliation pipeline that fetches, normalizes, matches, and flags discrepancies. Generates exception reports with root-cause suggestions.

Stack

Python, Apache Airflow, PostgreSQL, S3

Impact

Daily reconciliation: 6 hours manual to 18 minutes automated. Discrepancy detection accuracy improved to 99.2%.

Data Pipeline 2023

Multi-Tenant Reporting Infrastructure

A SaaS platform serving 200+ tenants had no standardized reporting. Each client request required ad-hoc queries and manual CSV exports.

Solution

Designed a multi-tenant data warehouse with tenant-isolated schemas, scheduled ETL pipelines, and a self-service report builder with role-based access control.

Stack

PHP, MySQL, Redis, ClickHouse

Impact

Custom report requests eliminated. Clients self-serve with 95% of previous use cases covered out of the box.

Approach

A methodology, not a hackathon

Every engagement follows a structured process. This ensures the solution addresses the actual problem — not the symptom.

01

Map the bottleneck

Before writing any code, I trace the full operational flow — data sources, handoff points, decision nodes, failure modes. The goal is to find where time, accuracy, or capacity is lost. Most problems are visible in the first 48 hours of observation.

02

Design for the system, not the feature

Solutions are architected for the operating environment — current load, expected growth, failure scenarios, team capability. A system that requires a dedicated team to maintain is a liability, not an asset. I design for operability from day one.

03

Build with observable defaults

Logging, metrics, and error tracking are not afterthoughts. Every system ships with structured logging, key performance indicators, and alerting thresholds. If it breaks, you know before the user reports it.

04

Validate against the original problem

Delivery isn't complete when the code deploys. It's complete when the measured outcome matches the defined target — processing time reduced, error rate lowered, manual hours eliminated. Numbers, not assumptions.

Trust Signals

Systems in production

Multinational Procurement Platform

Active

Dynamic approval engine processing 2,400+ approvals monthly across multiple business units and geographic regions.

Logistics Integration Gateway

Active

Centralized API gateway handling 23 external service integrations with automated failure detection and recovery.

Fintech Reconciliation Pipeline

Active

Daily automated reconciliation across 6 financial data sources with 99.2% discrepancy detection accuracy.

SaaS Multi-Tenant Reporting

Active

Self-service reporting infrastructure serving 200+ tenants with role-based access and automated data isolation.

Internal Workflow Automation Suite

Active

Suite of automation tools handling employee onboarding, data synchronization, and compliance reporting across 3 internal systems.

Start a Discussion

Have an operational bottleneck? Let's fix it.

I work best when there's a clear problem to solve. If you have a workflow that's too slow, a system that breaks too often, or a process that should be automated — let's talk.

Ascaliko