Phantom Trace: an offline AI that finds the ghosts in your CI/CD logs

Описание и применение

Phantom Trace – локальный инструмент для обнаружения повторяющихся ошибок в логах CI/CD, использующий открытый embedding‑модель для семантической кластеризации и отображения исходных строк.

Задача

Recurring CI/CD failures often reappear with different wording, making them hard to detect and prioritize.

Результат

In a sample run of nine fictional events, Phantom Trace produced three recurring clusters (sizes 3,3,2) and one isolated event, matching the intended pattern.

Рабочий процесс

• Paste timestamped CI/CD events into the local UI
• Phantom Trace converts each message to a semantic vector using all-MiniLM-L6-v2 via ONNX Runtime
• Vectors are mean‑pooled and L2‑normalised; cosine similarity is computed
• Pairs exceeding a threshold (default 0.55) are connected; connected components form incident clusters
• The medoid event of each cluster is chosen as the title; all supporting log lines are shown
• Users can adjust the similarity threshold with a slider to refine clustering

Доказательства

Phantom Trace is an offline incident‑archaeology tool for CI/CD teams. It uses an open‑weight sentence‑transformer to connect historical failures that mean the same thing even when their exact wording changes, then shows every source event behind each cluster.

Источник

The default threshold was 0.72. The UI looked polished, but a real model integration run exposed a problem: it found only one of the three intended recurring families. At 0.55, the model produced the intended 3, 3, 2, 1 cluster sizes.

Источник