I'm currently a backend intern working on CMS, back-office, and internal gift-card systems. The ticket descriptions say "feature work," but the internship has quietly taught me more than that: system architecture, performance tuning, database design, Git flow, and working inside an AI-assisted coding pipeline.
Outside work, I build small systems to understand how things actually work under the hood — not just how to call an API.
A small key-value store built to understand storage engines and performance trade-offs from the ground up.
Full-stack platform — course management, enrollment, progress tracking.
A multi-agent LLM system handling end-to-end sales conversations for e-commerce — my first real encounter with how unpredictable and costly LLM usage gets at scale once it's not a toy demo anymore.
Most of my recent reading and side-building has gone toward one specific thing: how systems track, price, and account for usage — idempotent event handling, concurrency-safe ledgers, webhook reliability. It started as curiosity from the AI agent project above (AI usage is expensive and hard to track accurately), and turned into the thing I now go deepest on outside work hours.
class MyApproach:
principles = [
"Understand the layer below before using the one above it",
"A system that can't prove what it did can't be trusted",
"Measure before claiming it works",
"Fundamentals first, specialization on top",
]
currently_learning = [
"Auth, API design, queues/workers, async & event-driven patterns",
"Idempotent event ingestion and concurrency-safe ledgers",
"Database performance and storage internals",
]Backend 
Systems & Data 
AI / LLM 
Domain
- UUIDv7: Why You Should Stop Using Auto-increment Integers and UUIDv4
- Analyzing Agentic Workflow Architecture via OOP: A loop.py Case Study
→ All posts on thaig2pro.github.io
