I'm a Java backend developer who enjoys the part of software most people avoid: what happens when two requests hit the same account at the same time, when a client retries a payment, or when a message gets delivered twice.

My projects are built around those problems: concurrency control, ACID transactions, idempotency, and event-driven architecture, with Spring Boot, PostgreSQL, Redis and Kafka. I finished my B.Tech in Computer Science in May 2026, and alongside my studies I've been working as an AI Response Evaluator, which has sharpened my eye for edge cases and precise documentation.

💬 My motto: "Make it correct first, then make it fast, and write down what you couldn't verify."

🔭 Working on   →  Distributed systems, ledgers and event-driven services
🌱 Learning     →  System design at scale, advanced Kafka patterns, cloud deployment
🤖 Also into    →  Local LLMs (Ollama), code generation, robotics with Arduino
📍 Based in     →  Bangalore, India
📫 Reach me     →  naikprashant837@gmail.com

Java 21 Spring Boot 3 PostgreSQL Redis Kafka Flyway JUnit 5

A payment backend where money never goes missing, even under concurrent load and duplicate requests.

• Double-entry ledger: every transfer is recorded as matching debit and credit entries, using BigDecimal and Flyway-versioned schemas.
• Safe balances: pessimistic row locks, optimistic locking (@Version) with bounded retries, and consistent lock ordering to prevent deadlocks between opposite transfers.
• Idempotency: request hashes, durable PostgreSQL records and Redis/Redisson reservation. A test with 20 concurrent callers sharing one key produced exactly one balance change.
• Transactional outbox: a scheduled poller publishes PAYMENT_SETTLED events to Kafka. The consumer deduplicates processed events, acknowledges manually and routes failures to a dead-letter topic.
• Verified: concurrent transfers, duplicate requests, outbox publication and backlog processing, all covered by JUnit 5 integration tests.

flowchart LR
    A[Transfer Request] --> B[Idempotency Check<br/>Postgres + Redis]
    B --> C[Locked Debit and Credit<br/>Double-Entry Ledger]
    C --> D[(Outbox Table)]
    D --> E[Scheduled Poller]
    E --> F{{Kafka: PAYMENT_SETTLED}}
    F --> G[Consumer<br/>Dedup, Manual Ack, DLQ]

Java FreeMarker FastAPI PostgreSQL Docker Ollama Groq Gemini

Write one YAML spec, get a runnable backend.

• One spec → full backend: a Java generator produces a FastAPI + SQLAlchemy + PostgreSQL + JWT service with a Dockerfile and Docker Compose files.
• Fails early: a validation pipeline (schema checks, typed mapping, cross-reference checks) rejects invalid specs before any code is written.
• LLM-assisted logic: 3 providers (Ollama, Groq, Gemini) write service-method logic from per-method prompts, with output cleaning, validation and up to 3 retries.
• Tested for real: black-box end-to-end scripts in Docker across 6 domains. The shop API passed 55 of 55 checks and the ticket tracker passed 50 of 56, with the remaining gaps documented.

Spring Boot Python Kafka Ollama YOLOv8 Whisper Arduino Docker Compose

A home robot split into independent services that talk only through Kafka.

• 4 services, 5 Kafka topics: a Spring Boot "brain" plus Python services for speech, vision and navigation, deployed with Docker Compose.
• Stateful orchestrator: tracks navigation, speech, vision and intended-mode states, pauses on the wake word, resumes safely, and classifies spoken commands into 3 intents using a local LLM.
• Perception stack: wake-word detection, Whisper transcription, face recognition, YOLOv8 object detection and an ultrasonic sensor that stops the robot at 25 cm.
• Honest status: tested end to end on hardware. Wheel motion is still unverified, and that's documented in the repo.

AI Data Trainer / Response Evaluator · Outlier AI (Remote) · Aug 2024 – Present

• Evaluate and score LLM responses for relevance, factual accuracy and instruction-following against task-specific guidelines.
• Identify edge cases and inconsistencies, document evaluation decisions, and apply consistent criteria across distributed workflows.

I'm looking for a Java backend / Spring Boot role where I can work on reliable, high-integrity systems. If you're hiring, building something interesting, or just want to talk about distributed systems, get in touch.