GhostTrace AI: A Local AI That Investigates Digital Ghosts

Description and use

GhostTrace AI is a local tool that reads fictional activity logs, scores anomalies with transparent rules, and produces spooky detective stories using the Qwen2.5 3B model via Ollama, with a fallback engine for offline use.

Problem

Develop a local AI tool that can analyze fictional computer activity logs, detect anomalies, and generate spooky detective stories without relying on cloud AI services.

Outcome

All 25 automated tests passed, and the demo video shows the tool detecting anomalies, explaining flags, and generating a spooky narrative. The local inference model eliminates cloud costs and enhances privacy.

Workflow

• 1. Read supported JSON, LOG, and TXT files from a selected sample data directory.
• 2. Apply transparent rule-based scoring to each event to detect anomalies and generate explanations.
• 3. Use the local Qwen2.5 3B model through Ollama to generate a spooky story based on flagged events.
• 4. Display a filterable investigation timeline of events and scores.
• 5. If the local model is unavailable, invoke the fallback engine to produce a procedural story.

Evidence

GhostTrace AI analyzes sample activity logs, identifies unusual-looking events using transparent rule-based scoring, and uses a locally running language model to turn those events into a story.

Source

I built GhostTrace AI using Python, Ollama, Qwen2.5 3B, Streamlit, and Pytest.

Source

The project includes automated tests for log parsing, anomaly scoring, sample cases, timeline filtering, and story-generation fallback. During development, the reported test suite passed all 25 tests.

Source

By running the open-weight Qwen2.5 3B model locally through Ollama, the application can process its investigation data on the user's machine. There are no per-request cloud AI charges, and developers can inspect the source code, modify prompts, and experiment with other supported models.

Source