Amman-shah/ecommerce-production-incident-postmortems

Autonomous E-Commerce Production Incident Post-Mortems (NexusOS v2.0)

Dataset Summary

This repository contains high-fidelity, meticulously graded post-mortem diagnostic samples mapping real-world infrastructure failures in e-commerce ecosystems (WordPress, WooCommerce, and core transaction layers) directly to expert root-cause analyses and isolated code execution patches.

The data is structured into optimized binary Parquet formats, making it immediately ready for direct pipeline injection, model fine-tuning, or analysis via pandas and Polars.

Data Generation Pipeline (NexusOS v2.0)

Data assets are orchestrated and ingested entirely using local consumer hardware running a zero-dependency Python standard library stack.

1. Ingestion: Tracks raw JSONL telemetry, maps complex system crashes, handles connection timeouts, and manages native file locks at the edge.
2. Sanitization: Raw logs are passed through a local Ollama inference loop running an LLM to dynamically audit, sanitize, and strip out private data identifiers (PII).
3. Serialization: Cleansed rows are instantly compiled into structured Parquet tables and pushed directly here to the Hugging Face hub.

Schema Structure

Each sample in the dataset contains:

• instruction: The raw system error context and triage request.
• response: The validated expert architectural patch and recovery playbook.
• quality_metrics: Multi-tier evaluation data including character token counts and confidence scoring.

---

Built with NexusOS v2.0 — Shifting AI evaluation away from generic synthetic boilerplate and onto real-world production chaos.