Blog
- The Traveling Salesman Problem Is NP-Hard. EAX Matched the Optimum on 24,978 Cities.
TSP is famously NP-hard, yet Edge Assembly Crossover has matched proven-optimal tours on maps of up to 24,978 cities — by evolving roads, not city lists.
- Three outliers, one root cause: fixing PSO, FPA and GSA
How three swarm solvers — flocking birds, pollinating flowers, Newton's gravity — were stranded at 150–250% median gap on the TSP benchmark, and how one shared fix brought them into the pack.
- The passkey is the account: replacing our auth provider with self-hosted WebAuthn
How teeline swapped a hosted sign-in page and OAuth buttons for one-tap passkeys, self-hosted sessions, and API keys you can actually manage — and the honest trade-offs of doing it at zero users.
- Fourier algorithm: how it came to be
A TSP solver that encodes a tour as a closed curve in the complex plane and decodes with a plain argsort, designed together with Claude Opus, starting from a 37-line Python prototype.
- How I exposed a TSP solver to AI agents via WebMCP
A worked example of the WebMCP declarative tool API: four hidden HTML forms, no server round-trip, tested live with Gemini through the WebMCP Inspector extension.