Model Tests
A Python harness that gives coding agents one TypeScript shopping-cart task and changes only the instruction suffix.
Senior software engineer · AI systems (RAG, agents) · remote
Newest first. Each card opens a write-up, with links to a live demo or the source code when there is one.
A Python harness that gives coding agents one TypeScript shopping-cart task and changes only the instruction suffix.
ASP.NET Core middleware that limits concurrent requests per rule. Past the limit, a request gets 429 at once or after a short wait.
One ASP.NET Core process runs an order state machine: Kafka carries commands and facts, and Cosmos DB stores the saga document.
A single-flight caller in C#: one in-flight Lazy per key, removed when the call finishes, with a newer entry left in place.
A single-flight caller: one factory per key while the call is in flight, a shared future for the waiters, and no stored result.
A cited-clause RAG over Argentine FX regulation: ask what a BCRA circular says and get a Comunicación “A” citation, or silencio.
A Python CLI that finds unprotected private keys, built spec-first with GitHub Spec Kit: spec, plan, and tasks before code.
A curated catalog of software-engineering and security skills for coding agents, with full-text search on a static site.
A feed-forward neural network with backpropagation, written in Python and NumPy without an ML framework, that predicts fuel efficiency.
A tour of neural-network models that repaint photos in the style of famous artworks.
Weather-station records from 1975 to 2015 turned into world map tiles of temperature deviations, with a Scala.js viewer.
An O(n log n) N-body galaxy simulation in Scala, running in the browser with Scala.js and on the JVM.
TensorFlow 1.13.2, the last release that runs on 32-bit (i686) Linux, with an installer for Debian 12 and notes for building it.
Scoring 41 Argentine administrations, 1853–2025, on macroeconomic and fiscal pressure indexes, with a Python notebook and a paper.