BNNR – a closed-loop pipeline for improving vision models
XAI-driven model improvement loop, but Weights & Biases already tracks experiments better.
AI persona-based behavioral testing for web apps. No test scripts. YAML-configured. Vision-powered.
Vision-based E2E tests survive UI refactors, but Claude-per-run cost adds up fast.
QA engineers, full-stack developers building web apps, teams struggling with flaky E2E tests.
Playwright · Cypress · Sauce Labs (cloud E2E)
The idea: test scripts break when markup changes. Vision-based tests break when the UX actually breaks.
Personas have attributes like technical comfort, patience, and role that shape navigation behavior. A "frustrated non-technical admin" navigates differently than a "power user developer."
pip install specterqa specterqa init specterqa run -p demo
Cost: ~$0.30-$3.00/run depending on journey length. Built-in budget caps. Requires Anthropic API key.Previously called GhostQA — we rebranded after discovering ghostqa.com exists as an AI testing company. Clean break, no confusion.
GitHub: https://github.com/SyncTek-LLC/specterqa
MIT license. Alpha (v0.3.0). Feedback welcome.
XAI-driven model improvement loop, but Weights & Biases already tracks experiments better.
Handwriting that AI agents can read via MCP — analog meets agent infrastructure.
Synthetic personas simulate attention heatmaps before you ship to real users.
Visualizes Claude Code rate limits locally without API calls or Node.js dependencies.
Daily arXiv scraping with Claude classification beats manual curation.
Guided authorization workflow beats writing Zanzibar schemas by hand.