InvestigateOS: Voice Investigation Simulator

InvestigateOS is a voice-driven investigation simulator designed for situations where understanding what happened requires more than asking an AI a question. Users enter a simulated ML production incident and investigate it naturally through voice. AssemblyAI provides the real-time voice interaction, while an investigation agent retrieves evidence, reasons over timelines, builds hypotheses, searches for counter-evidence, and explains its conclusions. Unlike a traditional voice chatbot or RAG system, InvestigateOS maintains a deterministic World State and Evidence Graph. The AI does not directly control the truth. It investigates evidence and proposes actions, while the simulation engine validates and executes state changes. For example, when an ML model's F1 score drops, the user can ask what changed before the degradation, inspect the evidence linking a feature-pipeline change to the incident, challenge the hypothesis with counter-evidence, and then say “Rollback the feature pipeline.” The simulated system changes state, generating new events and evidence that can be investigated again. The result is a closed-loop experience: Voice → Investigation → Evidence → Hypothesis → Action → State Change → New Evidence.