Many legacy or proprietary systems behave like blackboxes—hard to understand, yet mission-critical. I developed a Python-based tool using TensorFlow and LIME to reverse-engineer system behavior by analyzing input/output patterns.
The tool ingests logs, API calls, and network data to build predictive models using decision trees and random forests. Explainability is handled via LIME, generating readable insights for each model prediction. This approach was successfully used on a proprietary industrial IoT gateway to identify hidden failure modes without access to the source code or documentation.
