Open to Java backend and SDE roles

Backend engineer focused on reliable, event-driven systems.

At Accenture I build and support production Azure integrations in C#/.NET 8. I build Java 17 / Spring Boot systems of my own to go deeper on the same problems: retries, duplicates and failure recovery.

  • 47lost invoices traced to a root cause and fixed
  • 5,002files in the load test of that fix, with no overwrites
  • 3,700+blobs scanned by my duplicate-detection tool

About

Failure cases are the interesting part.

I'm an Associate Software Engineer at Accenture, working since November 2024 on event-driven integrations for a grocery-retail client, in C#/.NET 8 on Azure. I like problems that start with a symptom: a missing record, a message that arrives twice, a retry that isn't safe. I trace them to a cause, fix them, and check the fix with a test I can rerun.

Outside work I build Java 17 / Spring Boot services to practice the same ideas. I'm looking for Java backend and SDE roles on product teams, and I'm open to relocating.

Experience

Production work at Accenture.

Associate Software Engineer | Packaged App Development

Accenture · Chennai, India

– Present

Backend integrations for a grocery-retail client in C#/.NET 8 on Azure. This section is professional work only; the Java projects below are personal.

  • Two production integrations, owned end to end

    Customer events into a data lake and order documents into SAP S/4HANA, on Azure Functions. I own the ARM templates, Azure DevOps releases and production support.

    Measured in Application Insights over 30 days as request duration per invocation, not end-to-end delivery time: customer events 229 ms p95 / 365 ms p99; order documents 312 ms p50.

  • Order-document API on the Claim-Check pattern

    An HTTP API secured with OAuth2/JWT via Azure AD. The payload goes to Blob Storage and a reference to Service Bus, so ingestion doesn't wait on SAP. The XML-to-SAP mapping is covered by xUnit/Moq golden-file tests.

  • One function replaces a splitter and a timer

    A partner feed that ran as two functions is now one Service Bus-triggered function. Per-store idempotency (content-hashed, write-once Blob markers) lets retries skip payloads already sent, and permanent and transient errors are handled separately with Polly backoff and HTTP 429 Retry-After.

  • Alerting as code

    KQL rules deployed with ARM templates to three environments, de-duplicated by business key and routed to the team that owns the source system. I ran knowledge-transfer sessions on the triage workflow.

  • Duplicate detection for investigations

    A .NET tool that normalizes XML before hashing and flags repeated transaction and handling-unit IDs, used across 3,700+ blobs in Azure Blob Storage.

The two larger fixes are written up as case studies below.

  • C#/.NET 8
  • Azure Functions
  • Service Bus
  • Blob Storage
  • Application Insights
  • ARM templates
  • Azure DevOps
  • OAuth2/JWT
  • Polly
  • xUnit/Moq

Case studies

Two problems, from symptom to verified fix.

47 invoices lost to filename collisions

Problem
47 invoices went missing in an SAP FICO integration. An earlier mitigation (a millisecond in the filename, lower concurrency) didn't fully fix it.
Cause
Files processed in the same millisecond got identical blob names and overwrote each other.
Fix
Filenames now carry the invoice number and a 4-digit sub-second timestamp. Service Bus concurrency and Function App scale were capped to what the receiving system can absorb, trading some throughput for protecting it.
Verified
A load-test harness I wrote processed 5,002 files with no overwrites. The result applies to this fix only.
  • Azure Functions
  • Service Bus
  • Blob Storage
  • Load testing

Partner SFTP security change, zero downtime

Problem
A partner was retiring legacy SSH host-key types and ciphers on the SFTP endpoint used by five integrations across four Function Apps.
Cause
Partner-side security hardening, not a defect. Our connections already negotiated ssh-ed25519; the partner recommended rsa-sha2-512 or rsa-sha2-256.
Fix
Upgraded SSH.NET in those apps only and added a setting that puts host-key algorithms in priority order (rsa-sha2-512, rsa-sha2-256, ssh-ed25519) ahead of the library defaults. Nothing is removed, and the negotiated key is logged.
Verified
Tested in QA with the partner, who confirmed the files arrived. The production cutover had zero downtime.
  • SFTP
  • SSH.NET
  • Azure Functions
  • App configuration

Projects

Personal projects, built to learn in depth.

Flagship · Java 17, Spring Boot 3.2, Kafka, PostgreSQL, Redis

Reliable Webhook Delivery Platform

A multi-tenant webhook delivery platform in three Spring Boot services: ingestion, dispatcher and admin. It accepts events over HTTP and delivers signed webhooks to subscriber endpoints, with retries, a queryable dead-letter store and replay.

Webhook platform architecture A producer sends events to the ingestion service, which checks Redis and writes the event and an outbox row to PostgreSQL in one transaction. Events are published to Kafka fast and slow lanes, with a sweeper republishing unpublished rows. The dispatcher claims each event in an inbox table and delivers signed webhooks to the subscriber. Failures go to retry topics and then a PostgreSQL dead-letter store, which the admin service can replay through the outbox. sweeper replay via outbox ProducerHTTP request Ingestion serviceaccepts events Kafka fast lane · 4 threads slow lane · 1 thread DispatcherHMAC · circuit breaker Subscriberwebhook URL Redisdedup check PostgreSQLevent + outbox PostgreSQLinbox claim Retry topicsbackoff PostgreSQL DLQqueryable store Admin serviceendpoints · DLQ · replay 1 2 3 4 5 6 7 publish / deliver recovery / replay
  1. Intake. A Redis check drops duplicate requests.
  2. Record. The event and an outbox row commit in one PostgreSQL transaction.
  3. Publish. After the commit the event goes to Kafka; a FOR UPDATE SKIP LOCKED sweeper republishes rows left unpublished.
  4. Claim. An inbox row on (event_id, endpoint_id) with ON CONFLICT DO NOTHING.
  5. Deliver. An HMAC-SHA256-signed request, through a per-endpoint circuit breaker.
  6. Retry. Exponential backoff on separate topics, then a PostgreSQL dead-letter table.
  7. Replay. The admin service replays dead letters through the outbox.

Design decisions, and why

Outbox, not a direct publish

A service can't write to a database and a broker atomically. Committing the event and an outbox row together removes the gap where an event exists with no record that it must be published. The sweeper recovers rows never published.

Inbox claim, with an honest limit

Kafka can redeliver after a rebalance, so the claim recognizes repeats. Redis is a cheap first filter and the database constraint is the authority. This reduces duplicate webhooks but isn't exactly-once, because an HTTP call can't join a database transaction.

Lanes and circuit breakers

Degraded tenants run on a 1-thread lane and healthy ones on 4, and each endpoint has its own circuit breaker. Non-blocking retries give up strict FIFO, so ordering is best-effort per tenant.

Dead letters in PostgreSQL

They need to be queryable by tenant, mutable and kept beyond Kafka's retention. Replay goes through the same outbox path as new events.

Subscriber URLs are untrusted

Payloads are HMAC-SHA256 signed and admin calls need an API key. The SSRF check resolves the hostname at registration and rejects loopback, private, link-local, wildcard and multicast addresses, after normalizing hex and decimal hosts.

Tested at five layers

Unit, persistence, service, web and end-to-end tests with JUnit 5, Mockito, Testcontainers and WireMock. Recovery was exercised at two failure boundaries: after the database commit and after Kafka publication.

More projects

Full stack · Java

Blog-Zone

A blogging platform on a custom Spring Security filter chain, with role-based access, post CRUD, and like, comment, save and follow actions over AJAX. JPA/Hibernate on MySQL, containerized with Docker.

  • Spring Boot
  • Spring Security
  • JPA/Hibernate
  • MySQL
  • Docker

Full stack · Java

WeatherSpot

A Spring Boot weather app on the Visual Crossing API: current conditions, hourly and 15-day forecast views, and historical data back to 1973. It rotates between API keys with a fallback to stay within rate limits.

  • Spring Boot
  • Java 17
  • REST API
  • Thymeleaf
  • Docker

Developer tool

PushNPull

A lightweight API testing tool that handles GET, POST, PUT and DELETE requests with Basic Auth and Bearer tokens, and monitors response latency in real time.

  • Node.js
  • Express.js
  • REST APIs

Chrome extension

SimpleDefine

Look up a word's definition from the extension popup, or by double-tapping a word on any page.

  • JavaScript
  • Chrome extension

Skills

What I use, and where I've used it.

Used in production at Accenture

Platform

  • C#
  • .NET 8
  • Azure Functions
  • Service Bus
  • Blob Storage
  • Table Storage
  • Application Insights (KQL)
  • ARM templates

Delivery

  • Azure DevOps CI/CD
  • Git

Patterns and security

  • Claim-Check
  • Idempotency
  • Dead-lettering
  • Polly retries
  • OAuth2/JWT
  • SFTP (SSH.NET)

Testing

  • xUnit
  • Moq

Projects and training

Languages

  • Java 17
  • SQL
  • JavaScript
  • x64 assembly

Java stack

  • Spring Boot 3
  • Spring MVC
  • Spring Data JPA
  • Hibernate
  • Spring Security
  • Spring Kafka
  • Resilience4j
  • Maven

Data and messaging

  • Apache Kafka
  • PostgreSQL
  • Redis
  • MySQL

Testing and tools

  • JUnit 5
  • Mockito
  • Testcontainers
  • WireMock
  • Docker
  • Prometheus
  • Grafana
  • Postman
  • Swagger/OpenAPI

Also used

  • Node.js
  • Express.js

Fundamentals

  • Data structures and algorithms

Background

Achievements and education.

Algorithms

Solved 720+ LeetCode problems in Java and SQL, including 50+ in SQL.

Training

Completed Accenture's three-month Cloud-Native Microservices residency covering Java, Spring Boot and Docker. It was training, separate from my production work.

Mentoring

Taught Java, OOP and data-structures sessions as one of three senior mentors in a college coding bootcamp.

Education

Bachelor of Engineering in Computer Science and Technology

SNS College of Engineering, Coimbatore, India

– · CGPA 8.8 / 10

Contact

Let's talk.

I'm looking for Java backend and SDE roles, and I'm open to relocating. Based in Chennai, India.