Archaeologist: Oracle PL/SQL System Analyzer - Sell Project

Banks running old Oracle PL/SQL core systems spend months finding out by hand what depends on what before they can change or migrate anything. Archaeologist automates that analysis. It connects to Oracle with a read-only account, reads the database's own catalog of objects, dependencies and source code, parses the PL/SQL, and builds a knowledge graph in Postgres. Every edge in the graph records where it came from. Anything the tool can't prove is recorded as a flag, never as a guessed link. Everything runs on your own servers, and the source code never leaves the building. Features Extraction: 10 Oracle extractors covering objects, dependencies, source, triggers, constraints, scheduler jobs, synonyms, security grants, auditing and compile errors. PL/SQL parser: finds which procedures read and write which tables and which procedures call which, flags dynamic SQL, detects risky patterns such as WHEN OTHERS THEN NULL and commits inside loops, and records the units it can't parse. Analysis: column-level lineage, inferred foreign keys, change hotspots and code churn, impact analysis, security review, and snapshot comparisons between runs. Business-rule extraction: a local LLM (Ollama, qwen2.5-coder) turns PL/SQL logic into readable business rules, with a rule register. No cloud AI is needed. Web UI (React/TypeScript), 25+ screens: dashboard, graph explorer, impact analysis, read/write matrix, data model, flows, findings, hotspots, security, migration planning, generated documentation (English and Russian), and a DB health page. API: REST plus an MCP server, so AI assistants can query the graph directly. Exports: XLSX, DOCX, CSV, JSON and Markdown reports. Local setup: Docker Compose starts everything, including a seeded Oracle XE database with a synthetic core-banking schema and a golden file of expected results for testing. Tech stack Python 3.12 (FastAPI, SQLAlchemy, python-oracledb), PostgreSQL, React + TypeScript, Docker Compose, Ollama. Size and status About 100k lines of code and 327 passing automated tests. It's a working MVP, tested against the synthetic database, not a production deployment. Remaining work: The PL/SQL parser needs to ignore comments and string literals before it detects reads and writes. A few tests fail, and there are lint and type-check errors to clear. A separate operations module has its backend built but no UI yet. What's included Full source code, the architecture plan and phased build plan, test data and golden test files, Docker setup, and transfer of IP rights.