sandbox-execution

Gymnasium RL environment for training LLM agents to autonomously debug and fix Python code with secure sandboxing and test-driven feedback.

A multi-room collaborative IDE built with event-driven Socket.io state concurrency and isolated, secure sandbox execution environments.

A hybrid LLM routing agent built for the AMD Developer Hackathon. Optimizes token usage and costs by routing queries locally (Qwen 1.5B) or to remote APIs (Fireworks AI) inside an AST-secured sandbox. Includes a premium React visualizer dashboard and FastAPI backend.

ARC-AGI program synthesis: LLM-written solve() programs verified in a sandbox; 22/23 curated eval tasks

🔥 A LLM based API to source, prepare, analyze, and visualize any data.

A lightweight, zero-dependency, headless webhook relay and data transformation pipeline. Configure pipelines via YAML and map payloads with isolated JavaScript.

AI software engineering agent for translating, restructuring, compiling, executing, validating, and benchmarking multi-language codebases.

Ember CoreUI is the standalone, multimodal reference interface for the E.M.B.E.R. cognitive architecture. Completely isolated from the game backend of the Starlight Unit, the system offers its own infrastructure with DB, server stack and live browsing including a visible ghost cursor. Designed for autonomous, CPU-based cognitive research.

Streamlit app that answers natural-language questions about tabular data using a multi-agent Gemini pipeline: plan → generate Python → sandbox → charts & summaries