Research Tasks

Phase 1: Discovery & Scoping

• [ ] Compile list of primary sources (Debezium docs, MySQL CDC guides, enterprise blogs)


• [ ] Draft initial mental model of CDC-to-AI pipeline

Phase 2: CDC Landscape

• [X] Research Debezium MySQL connector: architecture, setup, connectors, limitations


• [X] Research Maxwell daemon for MySQL: pros/cons vs Debezium


• [X] Research other OSS MySQL CDC options (if any)


• [X] Research SQL Server CDC (built-in) + Debezium SQL Server connector feasibility


• [X] Survey Postgres logical replication (wal2json, pgoutput) for future phase


• [X] Brief Oracle options: GoldenGate, Debezium Oracle connector status, cost


• [X] For each DB: required DB privileges, operational risks, performance impact, offset storage, failure recovery


• [X] Write Report 1: CDC Technology Landscape


• [X] Tasks → decisions (recorded in .meta/decisions.md)

Phase 3: MVP Architecture

• [X] Define MVP scope boundaries (60 days, MySQL-only)


• [X] Choose CDC ingestion component (Debezium vs custom) — record decision


• [X] Design event normalization layer (raw → canonical schema)


• [X] Design semantic inference layer (canonical → business events)


• [X] Choose storage layer (append-only log + index: Postgres? SQLite? embedded?)


• [X] Choose AI layer (LLM integration: local vs cloud, summarization, search)


• [X] Design interfaces (CLI, optional web UI, MCP server later)


• [X] Draft deployment topology (containers vs binaries, network layout)


• [X] Write Report 2: MVP Architecture


• [X] Argue Build vs Buy (Debezium vs custom) — Deliverable 8

Phase 4: Event Model & Demo Domain

• [X] Define RawChangeEvent schema (fields from CDC)


• [X] Define CanonicalChangeEvent (normalized, DB-agnostic)


• [X] Define BusinessEvent (semantic, domain-specific)


• [X] Document handling of schema evolution, transaction boundaries, idempotency, ordering, multi-table correlation


• [X] Choose demo domain: Pharmacy chain (preferred) or Transportation


• [X] Design synthetic schema (branches, products, inventory, sales, adjustments, suppliers, users, roles, audit fields)


• [X] Plan synthetic data generator: realistic daily patterns + anomaly injection (fraud, stockouts, unusual adjustments)


• [X] Output SQL DDL (tables + indexes)


• [X] Write generator pseudocode (language-agnostic)


• [X] Write Report 3: Event Model & Demo Domain

Phase 5: Audit Intelligence Features

• [X] List 5–10 MVP features with immediate business value


• [X] For each feature: required data signals, simple heuristic baseline, AI improvement, pilot metrics


• [X] Write Report 4: MVP Features

Phase 6: MCP & Security

• [X] Design MCP server interface (tool definitions: search_events, get_daily_summary, get_anomalies, trace_entity, export_report)


• [X] Create security/compliance checklist (minimum privileges, network layout, encryption, audit logging, data minimization, LLM deployment choice)


• [X] Draft pilot proposal outline (emphasize low risk, quick value)


• [X] Write Report 5: MCP & Security

Phase 7: Implementation Plan

• [ ] Break MVP architecture into 25–40 engineering tasks


• [ ] Group tasks by milestone (e.g., Milestone 1: CDC ingestion; Milestone 2: Normalization; etc.)


• [ ] Define goal and acceptance criteria for each task


• [ ] Estimate difficulty: S (1–2 days), M (3–5 days), L (1+ weeks)


• [ ] Create Day 1–10 and Day 10–20 schedule (order tasks realistically)


• [ ] Write Report 6: Implementation Plan

Phase 8: Synthesis

• [X] Compile all reports into final package


• [X] Update index.md with summaries and status


• [X] Review decisions.md; ensure all major choices are logged


• [X] Prepare export for NotebookLM (optional)


• [X] Wrote FINAL_SUMMARY.md for handoff

Phase 9: Additional Reports (as needed)

• [X] Write Report 7: SQL Server & Oracle CDC Tooling Summary (standalone recap with detailed pros/cons, privileges, risks)

Phase 9: Additional Reports (as needed)

• [ ] Write Report 7: SQL Server & Oracle CDC Tooling Summary (standalone recap with detailed pros/cons, privileges, risks)