Blast Radius: Automated AST Dependency Mapping

Problem Statement In modern microservices and complex enterprise codebases, code changes are rarely isolated. Developers frequently modify internal utility functions, data models, or service helpers where local unit tests pass without warning. However, these changes regularly trigger silent regressions multiple hops downstream across consuming modules, payment gateways, or checkout pipelines. Traditional static analysis tools flag syntax or linting errors but lack architectural semantic context, while feeding entire codebases into large language models leads to prompt bloat, high token costs, security leaks of proprietary source code, and AI hallucinations. The Solution: Blast Radius Blast Radius is an automated code impact analysis tool that pairs deterministic AST (Abstract Syntax Tree) dependency graph extraction with IBM watsonx Granite reasoning. Instead of guessing relationships or scanning flat text, Blast Radius statically parses repository syntax trees to construct multi-hop downstream call graphs. It then feeds only the structured dependency topology into IBM watsonx Granite, delivering an automated, context-aware risk assessment and test recommendation matrix directly inside pull requests. Target Users & How They Interact Blast Radius serves enterprise software engineers, engineering managers, and QA/reliability architects through two touchpoints: Interactive Visualizer (Pre-Merge Exploration): Developers and architects run a local Streamlit dashboard where they select a target function. An interactive PyVis physics graph visualises the cascading blast radius, colour-coding the modified target (red), direct 1-hop callers (orange), and indirect downstream consumers (yellow) allowing engineers to inspect dependencies before opening a PR. Automated CI/CD Bot (Pull Request Audits): When a pull request is opened, a GitHub Action extracts modified symbols and computes the downstream blast radius. IBM watsonx Granite synthesizes this data into a standardized audit posted as a PR comment, outlining risk severity, affected features, required integration test suites, and preventative safeguards. What Makes It Creative & Unique Unlike purely generative AI tools that hallucinate call hierarchies or rigid linters that output cryptic log warnings, Blast Radius adopts a hybrid architectural pattern: deterministic structural parsing coupled with agentic semantic reasoning. Zero Code Exfiltration: Source code never leaves the organization’s secure boundary; only lightweight dependency schemas (symbol names and hop distances) are transmitted to the LLM. Deterministic Accuracy: The graph topology is derived mathematically from Python ASTs, eliminating hallucinated function calls. Contextual Prescriptions: watsonx Granite provides tailored recommendations—distinguishing between safe cosmetic changes and critical disruptions to billing or database layers. Effectiveness & Efficiency Architected and developed inside IBM Bob IDE across Ask, Plan, and Agent modes, Blast Radius drastically lowers the compute and token overhead of enterprise repository auditing. By replacing multi-thousand-token whole-file prompts with a compact JSON blast-radius payload, inference latency drops to under two seconds while maximizing watsonx token efficiency. It effectively shifts risk management left, preventing production outages before a single line of breaking code enters the main branch.