Spaces:

---

aaliyan1230

/

agent-probe

like 0

No application file

Hi HF team! I'm building AgentProbe — an open-source reliability testing and evaluation platform for AI agents. It stress-tests tool-calling agents under real-world fault conditions like schema drift, API timeouts, unstable networks, and adversarial tool responses.

This Space will host an interactive demo where users can run evaluations against popular agent frameworks (LangChain, LangGraph, smolagents) and see how they handle failure modes. The demo needs GPU access to run inference on open models (Llama, Mistral, Qwen) as the agents-under-test.

Related open-source work:

• tool-calling-reliability-benchmark
• turboquant-heretic-kv-bench
• false-refusal-suppression

I'm a Senior AI Software Engineer working on LLM evaluation and reliability research. A T4 GPU would be sufficient to run the demo. Thank you for considering!