Text-to-SQL on Oracle: two prompt rules took a light model from 35% to 64%

UniQL is a new benchmark that translates BIRD's 1,534 text-to-SQL questions into 16 SQL dialects. Its Oracle column is hard. The best published result is 49.61% (Claude Sonnet 4.5), and Gemini 2.5 Pro scores 26.14%. I ran schemagate on it with a light model, Gemini…

Before an AI migration, build a behaviour comparison harness

An AI-generated replacement can compile, start and produce plausible output while changing a business rule. For a migration, the useful question is whether you can identify and explain each observable difference from the old implementation. OpenAI’s October 19 research on low-resource programming languages distinguishes building…

Automating Large-Scale Number Enrichment: A Guide to Async Bulk Tasks

When your application needs to process tens of thousands of phone numbers, synchronous REST calls can quickly become a bottleneck. For large-scale enrichment—such as processing a list of 50,000 numbers—the carrier_batch service provides an asynchronous bulk workflow designed to handle high-volume data without blocking your…

AI agent observability: what to record and why

AI agent observability is the practice of recording what an agent tried, which evidence shaped each step, and what the user experienced, so you can debug failures, control cost and ship changes safely. It should answer more than "did the request finish?" It should help…

Selenium Interview Questions: Locators, Waits and Exceptions

Ten one-breath answers, each with a longer explanation and a Python example.Continue reading on Medium »

How to Package, Price, and Sell Domain-Specific Prompt Packs as Digital Products

A practical guide to turning structured prompt engineering workflows into recurring digital asset sales without relying on generic…Continue reading on AI & Side Hustles »

Run Claude Code with a Local Ollama Model: A Practical Guide

Want to try an AI coding agent without sending your project to a hosted model? This guide connects Claude Code to a local Ollama model. Commands are provided for Windows CMD, PowerShell, Linux and macOS. Our test setup: Windows, 8 GB RAM, Ollama, Claude Code…

OS Testing (Fresh code, no Linux distro)

Hey y'all! I just started making an OS with my own API's and boot kernel. No parent OS, just pure code and metal. If anyone is interested for using it/helping develop/find bugs, email me at rbxasas.research@gmail.com