Hey, I'm Gokul G — a backend engineer who builds systems that survive contact with the real world.

I've taken broadcasts from 8K to 100K recipients through festival-night traffic spikes, built lending platforms for UK building societies, and written a database engine from scratch. Java and Spring at the core, with AI wherever it earns its place.

Enterprise FinTech UK Building Societies platform at TCS

12.5x Scale 100K WhatsApp broadcasts & webhooks

40.77M MemTable writes/sec TangoDB • Maven Central

See my work Explore TangoDB Contact me

// About

The longer story

By day at TCS, I build and maintain the core loan origination platform for UK building societies using Spring Integration and Spring Batch — where correctness isn't optional and downtime is expensive.

Before this, at SKY X Digi Solutions, I led an engineering team of 3 to build a multi-tenant WhatsApp SaaS with an AI layer, scaling single-campaign broadcasts from 8K to 100K recipients (12.5x) while absorbing festival-season traffic bursts.

I like building systems that are fast, reliable, and useful to real people — whether that's off-heap storage engines in modern Java, fault-tolerant batch ETL pipelines, or wiring LLMs into pragmatic production tools.

// Flagship Open-Source

First open-source project: TangoDB

TangoDB v1.0.0 on Maven Central

Docs • GitHub

An embedded key-value storage engine, built from scratch in modern Java.

io.github.gokul-ceo:tangodb:1.0.0

40.77M ops/sec

Fast writes on MemTable

JMH, single-threaded, ~24.5 ns/op

CRC32C Checked

Crash-safe Write-Ahead Log

Replayed automatically on startup

Bloom + Sparse

Efficient reads & lookups

1% FPR skips disk lookups

Java FFM & Off-Heap

MemorySegment & Arena payloads

Eliminates JVM GC pressure • JDK 21+

Concepts behind TangoDB

LSM-Tree Design

Writes go to an in-memory MemTable, then flush to sorted files on disk (SSTables). Sequential writes are far faster than random ones, making this ideal for write-heavy workloads (same principle behind RocksDB and Cassandra).

Write-Ahead Log (WAL)

Before a write touches memory, it's appended to an append-only log. After an unexpected crash, replaying the log restores lost data. CRC32C checksums immediately detect half-written or corrupted records.

Bloom Filter

A compact probabilistic bitset answering "definitely not here" or "maybe here". With a 1% false-positive rate, 99% of queries for nonexistent keys skip disk I/O entirely.

Sparse Index

Indexes every Nth key. Binary search finds the target block in O(log N) time, followed by a tiny linear scan. Trades negligible scan overhead for an index that fits comfortably in RAM.

Java FFM & Off-Heap Memory

Uses the JDK 21+ Foreign Function & Memory (FFM) API (MemorySegment and Arena) to store binary record payloads natively off-heap, eliminating JVM Garbage Collection pressure.

Honest engineering note on benchmarks: The 40.77M ops/sec figure reflects single-threaded in-memory MemTable writes under JMH (~24.5 ns/op), not durable end-to-end disk writes. Technical precision builds trust.

// Experience

Work

Backend Java Developer 2024 — Present

Tata Consultancy Services (TCS)

Working on the backend of an enterprise loan origination platform for UK Building Societies, developed using Spring Integration and Spring Batch.

Built and optimized chunk-oriented batch pipelines for high-volume portfolio loan processing.
Integrated message-driven adapters and REST services to coordinate asynchronous transactional workflows across distributed banking services.

Developer • Team Lead (freelance) Prior to TCS

SKY X Digi Solutions

Led an engineering team of 3 to design and build a multi-tenant WhatsApp Cloud API SaaS platform with an integrated AI layer.

Scaled single-campaign broadcasts from 8K to 100K recipients (12.5x scale) • Led team of 3

Team Leadership: Led a team of 3 developers, guiding architecture, delivery, and code reviews while implementing high-throughput webhook pipelines.
Queue-based fan-out: Partitioned large 100K campaigns into queued worker jobs, keeping the API responsive while workers dispatched messages in parallel.
Rate limiting & backpressure: Dynamically regulated throughput to stay strictly within WhatsApp Cloud API boundaries without dropped requests or account penalties.
Fast acknowledge, process later: Returned immediate 200 OK responses to incoming webhooks to eliminate redundant retries, offloading payload processing to asynchronous queues.
Idempotency & spike absorption: Deduplicated messages by ID to prevent duplicate handling under heavy festival-season webhook bursts.

// Competencies

Tech Stack

Backend

Java, Spring Boot, Spring Integration, Spring Batch, REST APIs, Microservices

LLM integration, vision-based extraction, speech-to-structured-data, structured schema design

Systems

Embedded key-value storage (LSM-tree), off-heap memory (JDK 21+ FFM API), write-ahead logs, crash recovery

Integrations

WhatsApp Cloud API, webhooks, queue-based fan-out, rate limiting & backpressure

Hardware & Curious Jugaad

Scrappy, hands-on physical hacking: wiring microcontrollers, HX711 load cells, magnetic reed door sensors, and thermal printers to solve real-world friction.

// Horizon

Interests

Storage engines & systems internals

LSM-trees, WAL mechanisms, off-heap data structures (FFM), and memory-mapped I/O.

AI engineering

Wiring LLMs into production backend architectures, structured streaming, and latency optimization.

Embedded systems & curious jugaad

Microcontrollers, sensor telemetry, and scrappy physical hacks that let code touch the real world.

Robotics

Kinematics, mechanical actuation, and automation of repetitive physical workflows.

Investing: Capital allocation, unit economics, and compounding moats. • Indian political economy: DPI (India Stack), manufacturing policy, and supply chains.

// Connect

Where I add value

I engineer high-performance distributed systems built to handle scale, traffic, and complexity. My work spans resilient backend architectures, high-throughput data processing, event-driven systems, and performance-critical infrastructure. I also bring AI into production — integrating LLMs and intelligent capabilities into real-world systems without compromising reliability, scalability, or performance.

I enjoy tackling hard engineering problems, building systems from the ground up, and turning ambitious ideas into production-ready technology. If you're building something that needs to scale or pushing the boundaries of what AI-powered systems can do, let's talk.

Email me GitHub TangoDB Docs LinkedIn Twitter / X