DriveCore: AI-Powered AV Incident Response

DriveCore is a full-stack safety and engineering platform built to solve the "data deluge" problem in autonomous vehicle (AV) operations. When an AV experiences a GNSS failure, software regression, or unexpected road behavior, engineers typically spend significant time manually correlating logs, cross-referencing safety standards, and writing reports. Our platform automates this entire workflow using a sophisticated multi-step AI agent pipeline orchestrated with LangChain and powered by Qwen3. Running on AMD Instinct™ MI300X GPUs in the AMD Developer Cloud, Autopulse processes complex incident data through five specialized stages: Intake & Classification: Extracting critical telemetry like vehicle IDs, timestamps, and geolocations. Enrichment: Mapping failures to specific hardware and software subsystems (e.g., Localization, Perception, or Planning). Safety Logic Root Cause: Analyzing data against industry standards like ISO 26262 (Functional Safety) and SAE J3016 (Taxonomy of Driving Automation). Response Planning: Generating immediate mitigation strategies and operational recovery steps. Automated Reporting: Producing structured, professional incident documentation for regulatory and internal review. Beyond live incident response, Autopulse features a Branch Debug module for analyzing code diffs to catch software regressions and a Forensic Analysis tab for deep-dive security investigations and anomaly detection. By moving beyond simple RAG (Retrieval-Augmented Generation) to a reasoning-based agentic system, Autopulse acts as a force multiplier for AV safety teams, ensuring fleets stay safe, compliant, and on the road. This transition from manual triage to automated intelligence reduces response times from hours to seconds, providing the high-concurrency event processing required for modern autonomous fleets.