Payment Platform Engineering
Production backend work across payment-adjacent APIs, worker flows, integration boundaries, and reliability improvements, described at the domain level instead of naming sensitive internals.
Backend engineer from Malaysia · AI Researcher · Cloud Adoption DRI
Puvaan is a Malaysia-based backend engineer building payment systems, cloud infrastructure, developer tools, and privacy-first AI experiments.
Now
Working on payment-adjacent backend systems, driving AI and cloud adoption at Fiuu, and doing active AI research — reading papers, running experiments, and building personal AI projects. Most days run 12–14 hours. The work and the curiosity are the same thing.
How I Learn
I turn curiosities into tools: local AI assistants, OCR utilities, password/security labs, cloud experiments, and small products that teach the real tradeoffs behind production work.
Thread
My story is mentorship, self-learning, Apple TTP-related enablement, cloud systems, careful writing, and getting better at explaining complex work without leaking the work.
AI co-pilot pulse
1910M lifetime tokens across Claude Code and Codex — 586M in the last 30 days ending May 26.
Production focus
Payments
Current DRI
AI + Cloud
Awards
3
Favorite layer
Backend
Work story
Production backend work across payment-adjacent APIs, worker flows, integration boundaries, and reliability improvements, described at the domain level instead of naming sensitive internals.
Helping teams move from experiments to usable engineering practice: AI-assisted workflows, cloud adoption guidance, documentation, guardrails, and repeatable rollout habits.
Working around containerization, managed compute, serverless patterns, CI/CD, and infrastructure choices that make systems easier to deploy and operate.
Supporting payment enablement work at a safe public altitude, including Apple TTP-related delivery, integration readiness, and operational follow-through.
Small tools, parsers, dashboards, docs, and automation that shorten feedback loops for engineers and make complex systems easier to operate.
Privacy-first assistants, OCR workflows, post-quantum crypto demos, local LLMs, and applied research notes from building in public.
Selected work
2026
Split bills fairly in seconds with AI-powered OCR, multi-currency support, and a stateless privacy-first architecture.
2025
Transform physical documents into editable Word files using OCR and AI-assisted formatting.
2026
Generate passwords resistant to quantum computer attacks using ML-KEM, Argon2id, Go, Lambda, and WebAssembly.
2026
A Go CLI tool that tracks and attributes AI-generated code in git repositories.
2026
A production-grade AI personal assistant using Go, Next.js, Qdrant, AWS S3, Docker, and retrieval-augmented generation.
2026
Zero-trust serverless gateway that strips PII before requests reach public AI providers.
2026
Split bills fairly in seconds with AI-powered OCR, multi-currency support, and a stateless privacy-first architecture.
2025
Transform physical documents into editable Word files using OCR and AI-assisted formatting.
2026
Generate passwords resistant to quantum computer attacks using ML-KEM, Argon2id, Go, Lambda, and WebAssembly.
2026
A Go CLI tool that tracks and attributes AI-generated code in git repositories.
2026
A production-grade AI personal assistant using Go, Next.js, Qdrant, AWS S3, Docker, and retrieval-augmented generation.
2026
Zero-trust serverless gateway that strips PII before requests reach public AI providers.
Personal operating system
Notebook
Medium posts, local notes, applied research, and talk drafts in one place.
Research
AI systems
Talks
Docker Kuala Lumpur
Books
Matt Haig
Movies
Denis Villeneuve
Writing
Apr 18, 2026
A long-form engineering deep-dive into applying Karpathy's LLM-wiki pattern to Claude Code: transcript-replay auto-updates, Obsidian as source of truth, the honest trade-offs of Graphiti and custom builds, and a full survey of where else a personal knowledge base can live in 2026.
Jan 27, 2026
A technical deep-dive into building a stateless, multi-currency receipt splitter using Claude Vision OCR, Go Fiber, and React—with zero user data storage.