Systems that move money have no room for "mostly works." I build payment backends where every transaction is tracked from first sight to final settlement, and where a crash, a retry, or a chain reorg never counts the same deposit twice. Boring, careful, and correct is the whole point.
Most backend problems are really data problems. I spend my time on the layer underneath the API: schemas that stay honest as the product changes, queues that survive restarts, and queries that are still fast when a table hits its millionth row. TypeScript and Postgres are where I do my best work.
There's a wide gap between an impressive AI demo and a feature people rely on every day. I build on the far side of that gap: LLM-driven products with the retries, fallbacks, and guardrails to run unattended. The model is maybe a tenth of the work. The system around it is the rest.
I'm most useful when I own the whole problem: design, backend, frontend, deployment. Small teams don't have room for "not my part of the stack," and I like it that way. Hand me something vague and I'll come back with something running.
2026-07-13
Why we waited 32 blocks to credit Solana deposits
Solana has no mempool to watch and three commitment levels to choose from. Building a crypto payment handler, we credited deposits only at 'finalized', trading twelve seconds of delay for finality the consensus rules can't undo.
Samiul
Solana Commitment Finality Payments TowerBFT
2026-07-07
Why your first USDC send on Solana costs extra SOL
On Solana, token balances live in their own accounts, and accounts cost rent. Sending USDC to a wallet that's never held it means someone pays ~0.002 SOL to create storage. Here's who, and why.
Samiul
Solana USDC SPL Rent Fees
2026-07-05
Why your TRON USDT transfer silently failed
TRON splits fees into two independent resources: bandwidth and energy. A transfer can clear one and run out of the other mid-execution — here's what that looks like and how to avoid it.
Samiul
TRON USDT TRC-20 Fees
Samiul Alim - 2024
Dhaka, Bangladesh
