OraLift

Enterprise systems still run on decades of undocumented Oracle PL/SQL — packages nobody wants to touch because nobody's sure what they'll break. Generic AI code-review tools skim the surface; they don't reason about PL/SQL's stateful, transaction-heavy idioms. OraLift points at a folder of PL/SQL packages and produces four things: a call-dependency graph, a plain-English spec for every procedure and function, a generated REST scaffold (FastAPI routes + pytest stubs), and — the part most tools skip — a risk report naming exactly what can't be safely auto-migrated: dynamic SQL, autonomous transactions, package-level state, and DBMS_* calls. That risk report isn't an LLM guessing; it's four deterministic rules, so every flag is defensible and traceable to a specific line. We built this to lean directly into what IBM Bob is already best at: legacy and enterprise modernization. Rather than compete with the generic code-review bots this hackathon will be full of, OraLift targets an underserved niche — Oracle PL/SQL — grounded in real backend experience with the language. Bob was used throughout the 48-hour build: Architect Mode to plan each module's interface before writing code, Code Mode to implement extraction, risk-scanning, graphing, spec generation, and scaffolding. Stack: Python + FastAPI, a deliberately regex-based extractor (not a full grammar — a 48-hour tradeoff, not an oversight), networkx + Mermaid for the dependency graph, one Claude API call per unit for specs, Jinja2-generated REST/pytest scaffolds, and a Streamlit UI tying it all together.