Meridian

Built on large population modeling, Meridian instantiates your customer base as a field of autonomous algorithms and stress-tests marketing strategy variants against them iteratively until a winning case is achieved and a plan emerges.

Meridian is a strategy simulation platform that tests how different approaches play out before you commit to one.

You describe a scenario in plain text (a negotiation, a pitch, a campaign) and the system compiles it into a cast of actors with distinct personalities and a set of strategic variants. Each variant runs as its own isolated multi-agent dialogue simulation, with an LLM goal monitor checking progress turn by turn. When an attempt fails, a beam-search restart kicks in and carries lessons forward into the next run.

After all variants complete, a comparative synthesis ranks them and explains which approach held up and why.

A second quantitative layer runs Monte Carlo Tree Search across strategy branches, scoring each one on robustness against different externality scenarios to surface strategies that hold up under pressure, not just the best-case path.

Built with Next.js, Python FastAPI, Claude via the Anthropic API, Vultr, and Supabase. Simulation events stream to the frontend in real time over SSE, and results are saved to a run history for review and comparison.