Amy: Agentic Mathematical Engine - Looking for feedback & contributors

Community
1

I’m sharing an experimental project: Amy (Agentic Mathematical Engine) - an LLM-powered reasoning system for scientific computing in Julia.

What it does:

  • Takes natural language queries like “Design a Hohmann transfer from Earth to Mars”
  • Decomposes problems into goal DAGs via 5 specialized LLM agents
  • Executes tools across 7 mathematical domains (symbolic, numerical, tensor, coordinate, propulsion, matrix, physics)
  • Verifies results (units, conservation laws, bounds) and explains reasoning

Architecture highlights:

  • Fan-out/fan-in parallel execution for independent goals
  • Local Ollama/DeepSeek-R1 integration (no API keys needed)
  • Token budget management with circuit breakers
  • Built on Symbolics.jl, DifferentialEquations.jl, Unitful.jl, etc.

Repo: GitHub - agenisea/agentic-mathematical-engine: An intelligent Julia agent for scientific computing with natural language queries and multi-domain mathematical reasoning.

This is a WIP/experiment and has some bugs. Contributors welcome!

I’d love feedback on:

  • The agent orchestration approach
  • Tool design patterns for Julia
  • Use cases you’d want to try
1 Like