Skip to content
Alvaro Serero

Back to the home page

Python Maker Bot

LLM-driven Python code generator and sandboxed runner, in Rust

Period
Nov 2025 – Apr 2026
My part
Led the 4-person course version (90% of the code), then extended it solo.
Stack
  • Rust
  • tokio
  • reqwest
  • serde
  • axum
  • askama
  • HTMX
  • Docker
  • GitHub Actions
  • Ollama
Links
Source on GitHub

Built an LLM-driven Python code generator and sandboxed runner in Rust (8k lines, 117 tests, Docker sandbox, Axum/HTMX dashboard, from-scratch RAG). Led the 4-person course version (90% of the code).

  • Led a 4-person Rust project (ESILV, 2025) building an LLM-driven Python code generator and runner (tokio, reqwest, serde): 90% of the 1,700-line codebase and all 51 tests. Features: multi-turn refinement, dependency auto-install, and an interactive execution mode for GUI and input-driven scripts.
  • Extended it solo into an 8,000-line Rust tool: Docker sandbox with per-run virtual environments, ruff and bandit pre-flight checks, a multi-provider LLM client (Hugging Face, Ollama, OpenAI-compatible), an Axum + HTMX dashboard streaming logs over WebSocket, and a from-scratch RAG. 117 tests, clippy-clean, GitHub Actions CI.
  • Engineered API resilience (exponential backoff with jitter on 429/5xx, fail-fast on 4xx), execution timeouts, a TOML config chain and py_compile/traceback auto-refine loops for AI-generated scripts.

Honest framing

About 6,200 lines written by hand across both repos. The v0.4.0 streaming, /explain, /project, /session and CI features came from one reviewed AI-assisted PR.