Available for opportunities

Kavin S P

Full-Stack & DevOps Engineer

I work top to bottom of the stack — React and Next.js down through Node, Go and Java, into the Kubernetes clusters it all runs on. I gravitate toward the problems nobody's assigned yet: the database nobody questioned, the bill nobody optimized, the image nobody shrank.

$ whoami
full-stack engineer · devops · self-taught · 2.5 yrs in production_
01

About

I'm an engineer who moved into software from a mechanical engineering background, taught myself the stack, and have spent the last two and a half years shipping production systems for a travel-tech platform — flight and hotel booking, running across GDS and NDC integrations at real scale.

My work spans the full stack: building services in Node.js, Go and Java; building interfaces in React and Next.js; and running all of it on AWS EKS with Kubernetes, Jenkins, and Docker. I care less about which language a problem is in than about actually solving it — which is usually the thing that lets me pick up whatever's broken, regardless of which repo it lives in.

📍 Chennai, India ⚙ Java · Go · TypeScript · SQL ☁ AWS EKS · Docker · Jenkins
02

Departures — before / after

Terminal 1 · Performance & Cost — : — : —
Booking failure rate
30%
2%
−93%
Loyalty API (p95 latency)
22.8 s
2.39 s
−89%
Contract DB size (MD5 dedup)
19 GB
<1 GB
−95%
Monitoring spend (New Relic)
$2,000/mo
$200/mo
−90%
Next.js Docker image
500 MB
50 MB
−90%
OTP-UI Docker image
3.6 GB
1.6 GB
−56%
Seat map parser
8.5 s
3.5 s
−60%
All figures measured in production · New Relic / internal benchmarks
03

Selected work

Airline Contract Management System

Greenfield

Built end-to-end from scratch: versioned contract APIs, MD5-based deduplication that cut database size 95%, presigned S3 uploads, audit events and real-time updates via SSE.

Go FiberClickHouseReactgolang-migrate

Solo NDC Airline Integration

Ownership

Owned an end-to-end NDC integration for an international carrier as the sole developer — SOAP/XML request-response mapping, error handling, and certification against the airline's spec.

SOAP/XMLNDCJava

Kubernetes Reliability Work

Infra

Resolved intermittent 502s on EKS by fixing deregistration delay and adding pre-stop hooks; implemented container-aware dynamic heap sizing from cgroup v2 limits for Node and Java services.

EKSKubernetescgroup v2Node.js

Design System (Storybook)

DX

Fixed TypeScript declaration file generation that had stumped senior frontend engineers, restoring autocomplete for every consumer team, and introduced CSS code splitting to cut bundle output from 1.5MB.

StorybookChromaticvite-plugin-dts

Observability & Cost Overhaul

Cost

Rebuilt monitoring configuration across services, cutting New Relic spend 80% (~$19K/year) while improving signal quality with Grafana, Loki, and structured correlation-ID logging.

GrafanaNew RelicLokiWinston

Feature Flags: Code → DB + Redis

Platform

Replaced hardcoded feature flags with a DB and Redis-backed toggle system, turning what used to require a rebuild-and-redeploy into a zero-deploy change across the platform.

RedisPostgreSQLNode.js

K6 Load Testing Platform

Perf

Built a full booking-flow load testing platform simulating the entire flight-booking chain end to end. Used V8 profiling under production load to trace the real CPU bottleneck to XML parsing, and proved vertical scaling gave zero throughput gain before the team over-invested in bigger pods.

k6PrometheusLokiGrafana

AFKL NDC Cloud Migration

Solo

Led an Air France-KLM NDC integration migration from self-hosted infrastructure to Azure end-to-end — sandbox and production re-certification, mutual TLS with client certificates, GZIP-compressed SOAP payloads, and a PCI-DSS card tokenization proxy — while keeping every other airline integration backwards compatible.

SOAP/XMLmTLSAzure

Emergency Git Forensics

Recovery

Recovered roughly 20,000 lines of code lost to an accidental force-merge — 20 days and dozens of commits after the fact — extracting the lost work while preserving three weeks of legitimate history layered on top of it.

Git
04

Boarding pass — AI & LLM tooling

CLARITY AIR · AGENTIC CLASS E-TICKET
PassengerKavin S P
FromPrompt
ToProduction
CarrierClaude Code / CLI · OpenAI Codex
FlightMCP-020 (20+ managed, 2 self-built)
ClassRAG
GateAutomated Code Review
SeatAutonomous Incident Triage

Ships AI tooling, not just uses it: built two custom MCP servers from scratch — a Go log explorer (13 tools over production logs) and a TypeScript SQL gateway with write-protection — wired Claude CLI into Jenkins for automated PR review across the org, and built a daily pipeline that triages booking failures autonomously across SQL, Grafana, log, and Azure DevOps MCPs, from raw failure to a pre-analyzed bug ticket.

Boarding Always Seq 011
05

Manifest

Languages
Java, Go, TypeScript / JavaScript, SQL, Shell/Bash, XML/SOAP
Backend
Spring Boot, Spring WebFlux, Spring Security, Go Fiber, Node.js / Express, NestJS, AWS Lambda
Frontend
React, Next.js, Vue.js, Tailwind CSS, Redux/Zustand, TanStack Query, Storybook, Vite/Webpack
Databases
PostgreSQL, MySQL, ClickHouse, Redis / ElastiCache, Liquibase, golang-migrate
Cloud & infra
AWS (EKS, EC2, S3, Lambda, ECR, SQS, CloudFront, ELB/NLB), Docker, Kubernetes, Helm, Karpenter, KEDA
CI/CD
Jenkins, GitHub Actions, Azure DevOps, Maven, Gradle, SonarQube
Architecture
Microservices, multi-tenant systems, CQRS, pub/sub, DLQ, circuit breaker, feature flags, event-driven design
AI & LLM
Claude Code / CLI, OpenAI Codex, 20+ MCP servers managed (2 built solo in Go & TypeScript), automated AI code review (Jenkins + Azure DevOps), autonomous incident-triage pipelines, RAG (Qdrant, FastAPI, n8n), prompt engineering, AI-assisted commit generation
Domain
GDS (Sabre, Amadeus, Travelport, TBO), NDC (30+ airlines), EDI, SOAP/XML, flight & hotel booking lifecycle