Multi-region logistics at Amazon, banking onboarding at Accenture, workforce-management platforms at Ultimate Kronos Group — and lately, LLMs wired into the infrastructure I run. Shipped to users on two continents.
Featured System
Multi-region fault-tolerant microservices
logistics — us-east-1 · eu-west-1
Filter by///
Multi-region fault-tolerant microservices
Amazon · Logistics
Legacy compliance → service-oriented architecture
Amazon · Compliance / EPOD
RabbitMQ listener consolidation + API performance
Ultimate Kronos Group · Workforce management
Kubernetes MCP Server
Ultimate Kronos Group · AI & Tooling
Jira MCP Server
Ultimate Kronos Group · AI & Tooling
Salesforce MCP Server
Ultimate Kronos Group · AI & Tooling
Issue Context Agent
Ultimate Kronos Group · Agent · Confluence · Jira
Claude in the Support Workflow
Ultimate Kronos Group · Anthropic API · RAG
Claude Code skills — GitHub & hot-patch
Ultimate Kronos Group · Claude Code · Gradle · kubectl
AdResize AI
Side project · In development
ClientDesk AI
Side project · In development
Bluetooth-controlled Arduino car
College · IoT · 2021
// contact
Available for backend, distributed systems & AI engineering roles.
Region-local processing with replicated state — a regional failure degrades gracefully.Amazon · Logistics · 2024 – 2026
Multi-region fault-tolerant microservices
Owned design and operation of distributed services powering logistics workflows across regions, with fault-tolerant processing so a regional failure degrades gracefully instead of stopping deliveries.
~30% reliability improvement, fault-tolerant processing across regions.
Java
SQS / SNS
DynamoDB
ECS
CloudWatch
From one coupled monolith to independently deployable services with event contracts.Amazon · Compliance / EPOD
Legacy compliance → service-oriented architecture
Led the end-to-end migration of legacy compliance workflows into independently deployable services, and engineered the extensible APIs and data models behind the Electronic Proof of Delivery platform.
problem
Tightly coupled legacy flows, slow releases, heavy manual verification.
25–35% lower latency, 30–40% less manual verification, better maintainability and security.
Java
REST APIs
Kafka
Data modelling
IAM
One consumer service replaces N listeners — route, dedupe, retry in one place.Ultimate Kronos Group · Workforce management · 2026
RabbitMQ listener consolidation + API performance
Architected a support API that consolidates scattered RabbitMQ topic listeners into one well-observed entry point, and drove legacy API latency down through profiling and architectural fixes.
problem
Many independent listeners, duplicated handling logic, high operational overhead; slow legacy endpoints.
design
Single consumer service with routing by topic, shared error handling, PostgreSQL correction patches, query and caching optimisations.
result
Up to 40% lower API latency, reduced operational complexity, more efficient message processing.
Java
Spring Boot
RabbitMQ
PostgreSQL
Redis
claude — kubernetes-mcp› anything unhealthy in payments?▶ get_pods{"namespace":"payments"} 12 pods · 11 Running · 1 CrashLoopBackOff▶ tail_logs{"pod":"payments-worker-5c9","lines":200} HikariPool-1 — connection is not available…✓ pool exhaustion after 10:40 deploy · rollback (approve?)Read-only tools by default; mutating actions require explicit approval.Ultimate Kronos Group · MCP · Java · kubectl
Kubernetes MCP Server
Exposes cluster operations as MCP tools — list pods, tail logs, describe deployments, check rollout status — so Claude can debug a namespace conversationally, with read-only defaults and explicit approval for mutating actions.
get_pods
tail_logs
describe
rollout_status
approval gate
claude — jira-mcp› what is blocked in the current sprint?▶ search_jql{"jql":"sprint in openSprints() AND status = Blocked"}PAY-1187 retry storm on refund webhook P2PAY-1203 flaky integration test · onboarding P3▶ comment{"issue":"PAY-1187","body":"root-cause draft…"}✓ 2 blocked issues · summary posted to the sprint threadJQL search, ticket history, transitions and comments as typed tools.Ultimate Kronos Group · MCP · REST · OAuth
Jira MCP Server
Tools for searching issues with JQL, reading ticket history, creating and transitioning issues, and posting comments — letting Claude summarise a sprint, draft tickets from a bug report, or link incidents to work items.
search_jql
create_issue
transition
comment
claude — salesforce-mcp› open high-priority cases for Acme?▶ soql_query{"soql":"SELECT Id, Subject, Status FROM CaseWHERE Account.Name = 'Acme' AND Priority = 'High'"}00071234 Timesheet sync delayed Working00071301 SSO login loop (EU tenant) Escalated✓ 2 cases · fields limited by allow-listField-level allow-lists keep the rest of the org out of the model's reach.Ultimate Kronos Group · MCP · SOQL · REST
Salesforce MCP Server
SOQL query and record tools over accounts, cases and opportunities, with field-level allow-lists — so support and account context can be pulled into Claude without exposing the whole org.
soql_query
get_case
update_record
allow-lists
issue-context-agent — INC-4821issue INC-4821 · "duplicate payroll notifications"▶ plan runbooks → similar tickets → recent deploys▶ search confluence: 3 pages · jira: 5 linked · git: 2 commits▶ rank top: "Notification retry design" (0.91)✓ brief ready · 6 citations · est. 40 min savedPlan → search → rank → cite: the engineer starts with the full picture.Ultimate Kronos Group · Agent · Confluence · Jira · RAG
Issue Context Agent
Takes an incoming issue and goes looking for everything related: Confluence runbooks and design pages, similar or linked Jira tickets, past incidents and code references — then returns a ranked context brief with citations, so the engineer starts with the full picture instead of a blank ticket.
plan → search → rank
Confluence
Jira
citations
support-triage — ticket #58213ticket "Approvals page times out for managers"▶ triage severity: P2 · area: approvals-api · confidence 0.86▶ hints 3 similar tickets · runbook: "slow approvals query"▶ draft customer reply prepared (214 words)■ awaiting engineer approval — nothing sent yetTriage, hints and a drafted reply — nothing leaves without an engineer.Ultimate Kronos Group · Anthropic API · RAG · tickets
Claude in the Support Workflow
Claude in the ticket pipeline: triage and severity suggestion, root-cause hints from logs and past tickets, and drafted customer replies — grounded on runbooks via retrieval, with an engineer approving before anything is sent.
triage
log analysis
runbook RAG
human-in-the-loop
claude — /wfc-jar-patch people-impl› /wfc-jar-patch people-impl▶ gradle clean build -x test BUILD SUCCESSFUL1m 12s▶ discover **/build/libs/ people-impl-9.4.1.jar(matches 9.4.x)▶ backup WEB-INF/lib/people-impl-9.4.0.jar → .bak▶ kubectl cp → wfc-backend-0:/tomcat/webapps/wfc/WEB-INF/lib/✓ patched on k8s pod + HA machine · rollback: /wfc-jar-patch --rollbackOne command replaces gradle → find the JAR → back up → kubectl cp → kubectl exec — with a rollback that restores the previous JAR.Ultimate Kronos Group · Claude Code skills · Gradle · kubectl · Tomcat
Claude Code skills — GitHub & JAR hot-patch
Two Claude Code skills written for the team. A GitHub workflow skill drives the day-to-day flow from the CLI — branches, pull requests, reviews and release notes with the team's conventions built in. A hot-patch skill turns local testing against real environments into a single command: it runs the Gradle build, discovers the freshly built JARs, matches version suffixes against what is deployed, backs up the current JAR and swaps the new one into the running backend's Tomcat WEB-INF/lib — on Kubernetes pods and on the HA (high-availability) machines — then rolls back on request.
before
A 15-step manual sequence of gradle, kubectl cp and kubectl exec, easy to get wrong and hard to undo.
after
One command with guard rails: confirmation of the JARs to patch, oldest/latest backups kept, and --rollback to restore the previous state.
why it matters
Engineers test a fix in a live pod in seconds instead of waiting for a full image build and deploy.
Claude Code skills
Gradle
kubectl cp / exec
Tomcat
GitHub CLI
rollback
Side project · In development
AdResize AI
Upload one ad creative, get back every platform size — Meta, Google, LinkedIn, YouTube, Stories — with the layout intelligently re-composed instead of dumb-cropped. Optional AI edits: swap background, change copy, localise text.
pipeline
upload → vision (detect subject / text / logo) → layout plan per ratio → outpaint + re-compose → QA check → zip of sizes
design
Async job pipeline on SQS + workers, each size rendered idempotently. The LLM produces a structured layout plan (JSON schema) — the renderer, not the model, draws pixels. A QA agent flags cut-off text and off-brand colours.
Java · Spring Boot
Vision LLM
Image Gen API
AWS S3 · SQS
React
Side project · In development
ClientDesk AI
A client-management workspace for freelancers and agencies. Drop in notes, call transcripts, briefs and past work; it drafts the proposal, generates a branded PDF and audits scope, risks and missing requirements.
pipeline
project intake → RAG over client docs → proposal agent → audit agent → PDF render → share / track
design
Per-client vector store so every draft is grounded with citations. Proposal, pricing and audit are separate agents sharing a typed project schema. Event-driven: upload → embed → summarise → notify.
Java · Spring Boot
RAG
PostgreSQL + pgvector
PDF generation
AWS
First run, 2021 — Arduino UNO, HC-05 Bluetooth module, motor driver, phone as the controller.College · IoT · 2021 · Built
Bluetooth-controlled Arduino car
First IoT build: an Arduino UNO motor car driven over Bluetooth from a phone — motor driver, HC-05 module, custom control sketch. Where the hardware-to-software curiosity started.
