Expiry Radar

Problem & Challenge: Modern software development relies heavily on external dependencies, rapid iteration, and temporary workarounds. While "technical debt" is often treated as a vague concept, a large portion of it has an exact, predictable expiration date. This "ticking code" include end-of-life (EOL) dates for runtimes like Python 3.10, the sunsetting of specific APIs (such as the massive deprecations in IBM Qiskit 1.0), expiring SSL certificates, and developers leaving comments like "TODO: remove this temporary hack by 2025." When these dates inevitably arrive, CI/CD pipelines break, security vulnerabilities emerge, and production applications crash, usually resulting in a midnight incident. Tracking these hidden expiration dates across a massive, evolving codebase is practically impossible for human engineers to do manually. The Solution: Expiry Radar is an autonomous, predictive maintenance tool that flips technical debt management from reactive firefighting to proactive, automated prevention. It identifies hidden expirations before they trigger, giving teams months of lead time instead of minutes. Target Users & Interaction: Expiry Radar is designed for Software Engineers, DevOps teams, and Engineering Managers. Users interact with the system in two primary ways. First, a Streamlit dashboard provides a clear, visual calendar timeline of all upcoming breakages, ranked by severity. Second, the system integrates directly into the developer workflow, automatically opening Pull Requests to fix the nearest expiring items before they break. Creative & Unique Approach: What makes Expiry Radar truly unique is our hybrid architecture, inspired by IBM's own Workflow design. Instead of blindly passing an entire repository to an LLM, which is slow and prone to hallucination, we use lightning-fast deterministic Python collectors. These scripts scan for hardcoded dates, dated TODOs, and dependency versions. We then feed only those precise, highly relevant snippets into parallel IBM Bob AI subagents. This minimizes token usage, cuts execution time to minutes, and prevents AI hallucinations. To enforce absolute accuracy, we implemented a strict "evidence rule": every single finding the AI produces must carry a specific file path, line number, and a URL to the source documentation (like an official changelog). Efficiency & Effectiveness: To prove our solution efficiently addresses the issue, we performed historical backtesting on a 2023 snapshot of the qiskit-machine-learning repository. We ran Expiry Radar as if it were 2023 and successfully predicted the exact breakages that occurred when Qiskit 1.0 launched, achieving an average of 7 months early warning. By pairing cheap deterministic scanning with advanced AI reasoning, Expiry Radar offers an incredibly effective way to neutralize ticking code.