Java backend · distributed systems · production engineering

I build backend systems that stay correct under load and failure.

Software Engineer with 4+ years across fraud screening, financial workflows and integration systems. I build services, tune hot paths, reason about transaction boundaries, and stay close enough to production to understand how the system actually fails.

Performance−40% p95

JFR-guided work across Hibernate access patterns, SQL, composite indexes, HikariCP and Redis caching.

production

Throughput+15%

Improved backend throughput after service, persistence and cache-path optimization.

production

Reliability99.8% uptime

Supported with CloudWatch, structured logging, alerting, runbooks, incident triage and RCA-driven fixes.

production

Operations35 → 12 min

Reduced MTTR by strengthening failure isolation, debugging paths and operational playbooks.

production

01 / Production systems

Production systems where correctness and failure handling were part of the feature.

Three systems from professional work, with proprietary implementation details intentionally generalized.

Fraud screening

Decisioning under retries, bursts and dependency failure

Stateless Java/Spring Boot service with idempotent APIs, Redis risk lookups and velocity counters, token-bucket rate limiting, async processing, circuit breakers, fallbacks and external risk-provider integrations.

Idempotency · Redis · rate limiting · async · circuit breakers · observability

Millions/daytransaction evaluations

99.8%uptime

35 → 12MTTR min

Case study ↗

Financial workflow

Transaction boundaries before CRUD convenience

Spring/JSP transaction and approval workflows using @Transactional, MySQL isolation levels, optimistic locking, SQL tuning, indexing, cursor pagination and Redis caching.

ACID · isolation · optimistic locking · SQL tuning · pagination · caching

ACIDconsistency

Optimisticconcurrency

Case study ↗

Integration platform

Keep one slow provider from slowing the product

React + Spring Boot + MySQL portal with circuit breakers, retry/backoff, parallel fetch, batch inserts, caching, database-write optimization and explicit failure recovery around external APIs.

parallel fetch · bounded retry · batch persistence · cache · provider isolation

−55%refresh latency

+33%engagement

Case study ↗

02 / Public systems

Public systems with inspectable code, tests and engineering constraints.

Each entry points to code a reviewer can inspect. Vortex is the flagship systems project; the remaining repositories show distributed-systems, Java, AI-tooling and MLOps breadth.

GPU systems · public

Vortex CUDA

An exact vector-search engine built from CUDA kernels through C++20, Java JNI and a Spring Boot API. The system keeps indexes resident on the GPU, supports FP16 storage with FP32 accumulation, and includes a reproducible benchmark harness instead of relying on one-off timing claims.

CUDA · C++20 · JNI · Java 17 · Spring Boot · exact Top-K · FP16

Read engineering notes ↗Repository ↗

16.7×lower P50 vs project scalar CPU reference

1.67–1.72 msFP32 P50 · 500K × 128 · Top-10

1.64K+FP32 QPS · batch 32

50%vector-storage reduction with FP16

distrib-txn-db

Java workshop implementation progressing through HLC, MVCC, distributed routing, transaction records, write intents, clock-uncertainty read restart and simplified serializable conflict prevention.

Java · HLC · MVCC · transactions 02

SpringMind-AI

Terminal-first Java coding assistant with an LLM tool loop for file reads/writes, shell execution and iterative task completion.

Java 21 · Maven · LLM tools 03

Aegis Insurance Decisioning

Java + Angular decisioning system for policy, claims, premium, risk and audit workflows. The public version explicitly uses an in-memory store to keep local review simple.

Java 17 · Spring Boot · Angular 04

JobAgent-AI

Public fork of MadsLorentzen/ai-job-search, used to explore agent-assisted job evaluation, document tailoring and interview workflows.

Fork · agent workflows · Claude Code 05

networksecurity

Learning-oriented end-to-end MLOps project with ETL, MongoDB, MLflow, Docker, GitHub Actions and AWS deployment components.

Python · MLOps · AWS · MongoDB

03 / Upstream engineering

Upstream fixes at failure boundaries.

The signal is not diff size. It is whether the failure mode is real, the fix belongs at the right boundary and the regression test proves it.

AutoMQ#3493 · Merged

Skip AutoBalancer metrics reporter on controller-only nodes

Process-role handling prevented broker listener resolution and reporter startup where the broker role was absent, with regression coverage across role combinations.

runtime / roles AutoMQ#3542 · Merged

Remove redundant deleted-topic cleanup notifications

Removed duplicate deleted-topic traversal in ElasticReplicaManager because BrokerMetadataPublisher already notifies the group coordinator after metadata deltas apply. Preserved local partition deletion and state-transition callbacks, with targeted unit-test coverage.

metadata / callbacks Apache Fluss#4263 · Open

Make cached RPC connections endpoint-aware

Connection identity moved from logical UID alone to UID + host + port so metadata endpoint changes cannot silently keep routing through a stale physical connection.

RPC / lifecycle Trino#30973 · Open

Prevent query failure after transaction commit starts

Coordinates commit and failure ownership so unrelated failures cannot override the outcome after an autocommit transaction begins committing.

transactions / race

04 / Career record

From backend delivery to production ownership.

The progression below focuses on systems shipped, production responsibility and measurable engineering outcomes.

APR 2025 — NOW

Linux Socials

Software Engineer

Designed and owned fraud-screening backend services, tuned latency and throughput, hardened dependency behavior, and supported production through observability, incident response, runbooks and RCA.

DEC 2021 — APR 2024

Merkle Inspire Pvt Ltd

Associate Software Engineer

Built Java/Spring financial microservices, transaction workflows and an external-data job portal; supported three production microservices and improved release velocity through testing, CI, reviews and cross-functional delivery.

JUN 2021 — NOV 2021

Merkle Inspire Pvt Ltd

Technical Intern

Implemented backend/API components and reduced query execution time by 27% through SQL rewrites and index tuning while building testing and database-performance fundamentals.

05 / Engineering surface

Engineering depth I can defend.

Concrete examples from production work and public systems across service design, data consistency, distributed behavior and operations.

Service design

Java · Spring Boot · Spring MVC · REST APIs · JPA/Hibernate · validation · exception handling · service boundaries

Data & consistency

MySQL · PostgreSQL · Redis · SQL tuning · composite indexes · transaction management · isolation · optimistic locking

Distributed behavior

Kafka · idempotency · retries · timeouts · circuit breakers · rate limiting · asynchronous processing · concurrency

Production

AWS · EKS · Docker · Kubernetes · Jenkins · CloudWatch · structured logging · alerting · runbooks · RCA · JFR

06 / Contact

Backend systems. Measurable behavior. Fewer surprises in production.

Open to backend and software-engineering roles where system design, performance and operational ownership matter.

pranayp.kadu@gmail.com ↗github.com/BackendArchitectX ↗