Skip to main content

What is AIMon

AIMon helps developers build robust and reliable generative AI applications. AIMon provides a suite of detectors that help detect issues like hallucinations, adherence of the generated text to the provided LLM instructions, conciseness and completeness of the generated text, and more. AIMon is agnostic to the underlying generative model and can be used both in offline evaluations, continuous monitoring and some detectors can be used as a guardrail in production systems.

What exactly does AIMon provide?

AIMon's platform includes detectors, a user-friendly SDK and a UI that aids in monitoring of offline evaluation and production systems. Here is a quick overview of the detectors:

  • Hallucination Detector: Given a context, query and a generated text, it detects if the output of a generative model was not generated from the provided context. This detector operates at low latency and has accuracy comparable to LLM based judging approaches.
  • Conciseness Detector: Given a context, query and a generated text, it detects if the output of a generative model is concise and does not contain unnecessary information.
  • Completeness Detector: Given a context, query and a generated text, it detects if the output of a generative model is complete and does not miss any important information.
  • Adherence Detector: Given a context, query and a generated text, it detects if the output of a generative model adheres to the provided instructions.
  • Toxicity Detector: Given a generated text, it detects if the text contains toxic content. We have a taxonomy of toxic content that includes hate speech, obscenity, threatening language, profanity, etc.

Reference Architecture of a Generative AI Application using AIMon

The figure below shows a typical generative AI application architecture that uses AIMon's hallucination detector.

aimon_llm_architecture_halldet_example

What's next?