Aran ShayanAran Shayan

One platform for banks and telecom operators to design, secure, publish and monitor their internal and external APIs.

New APIs are built without code, by dragging and connecting nodes into a flow, then protected with OAuth2 and JWT and deployed on cloud, on premise or both.

One app for where a developer’s time, traffic and disk space go: project and screen time, internet usage per app and network, system health and cleanup, Jira and team chat, Git delivery and a full API workspace.

It runs on Windows, Linux and macOS, keeps every byte on the machine, and ships as one Python file with small native helpers in Rust and Go.

Who may see which document, in which building, for which team? I designed a control plane for relationship based authorization: analysts model, test, version and publish permission schemas without touching production code.

Live decisions stay in the authorization engine; the designer owns the lifecycle around it, from draft to validated, reviewed and immutable published versions, with a test harness that explains every allow and every deny.

Every branch of a bank holds and moves cash every day. This system collects each branch’s daily cash position and the transfers between branches, reconciles them and reports to the central bank.

Head office sees liquidity across the whole network in one place, and discrepancies surface the same day instead of at the end of the month. In production at a national bank.

Before and alongside product work I designed and built company websites and online stores for small businesses and brands, from the first sketch to launch and the updates after it.

Sites that load fast on a phone, rank in search and are easy for the owner to update; stores with catalog, search, cart, checkout with payment gateways, and a panel to run orders and stock.

Recommenders work from history, and a first time traveller has none. Among about 792,000 candidate places, I built a meta-learning recommender that adapts to a new user from just a handful of check-ins.

Adding geographic and social context lifted NDCG@10 by 21.7% and Recall@10 by 26.3% over a graph neural network baseline, and a gradient-free prototypical network matched MAML while training about 3.5x faster.