AI for PostgreSQL DBA
Detailed Curriculum

Exact prompt patterns for PostgreSQL troubleshooting, query analysis, and configuration questions. Build a personal prompt library you keep and reuse forever.

Autocomplete stored procedures, generate migration scripts from plain-English descriptions, and code-review your SQL before it hits production — all inside VS Code.

Paste any EXPLAIN ANALYZE output — get plain-English diagnosis, bottleneck identification, and index recommendation in seconds. Includes prompt templates that actually work.

Feed slow queries to AI, get rewritten versions with justification. Learn which suggestions to trust — and when not to. Hallucination awareness from day one.

Systematically build, version, and organize reusable prompts for the 20 most common PostgreSQL production problems. Your personal AI runbook — structured, searchable, reusable.

Turn a 2-hour investigation into a 5-minute diagnosis. Exact prompt structures that make ChatGPT a reliable first-responder for PostgreSQL incidents — locks, slow queries, OOM events.

Automatically generate step-by-step operational runbooks from incident history. Never write the same runbook twice. Templates for failover, bloat remediation, and connection storms.

Feed PostgreSQL log files to the ChatGPT API. Get a structured daily report: CRITICAL / WARNING / INFO ranked by severity. Python + OpenAI — beginner friendly, production ready.

Connect to live pg_stat_bgwriter, pg_stat_activity, pg_stat_user_tables. Feed data to AI. Generate executive-readable weekly health summaries automatically every Monday.

After every incident, AI drafts the full post-mortem — timeline, root cause, impact, remediation. Reduce post-incident documentation from 2 hours to 5 minutes.

Every resolved incident is stored as an embedding in pgvector. When a new incident occurs, the system finds the 3 most similar past incidents and surfaces the proven fix automatically.

Automate weekly ingestion of closed incidents into the vector database. Your AI system gets smarter every week — without any manual work. Fully scheduled via Python + cron.

Generate weekly AI vs human fix comparison metrics. Show the value: how AI-assisted diagnosis cuts mean time to resolution by ~40% — with real numbers from your own incident data.

Parse PostgreSQL logs into structured chunks, generate embeddings, store in pgvector. Search logs with meaning — not grep. Find similar past errors instantly across months of history.

Connect Prometheus metrics to an LLM layer. When Grafana alerts fire, AI automatically correlates the alert with historical similar events and generates a triage summary — before a human looks.

A full pipeline — not just a prompt. Logs → parser → embedding search → LLM synthesis → formatted HTML report. Production-grade, repeatable, not a one-off script.

Before any schema change hits production — AI reviews it for missing indexes, unsafe lock operations (ACCESS EXCLUSIVE), sequence gaps, and performance risks. Integrated into CI/CD.

Automatically check schema compatibility, index parity, and sequence drift between Blue and Green environments before cutover. Built on real AWS RDS Blue-Green production patterns.

Understand Model Context Protocol — what it is, why it matters, and why it separates modern Staff DBAs from the rest. Connect Claude directly to live PostgreSQL metadata.

Engineers ask "why is the database slow?" in plain English. The MCP assistant queries pg_stat_activity, pg_stat_bgwriter, and lock tables — answers back in natural language. No SQL from the caller.

Build agents that autonomously query, diagnose, and report on database health. Design the human-in-the-loop boundary — when to auto-remediate vs escalate. A Staff-level engineering decision.

Chunking strategy for PostgreSQL logs and docs. Embedding model selection — latency vs cost vs quality. IVFFLAT vs HNSW at scale. Vector DB sizing for 100K+ incidents/month.

Design a RAG system serving multiple DB teams from one platform. Tenant isolation with RLS, access control, audit trails, and AI governance — organisation-scale design.

Why RAG answers degrade over time and how to detect it from the database side. Measure retrieval quality with PostgreSQL window functions — no Python needed.

ML-based detection of lock storms, bloat spikes, replication lag, and connection surges — before they cause outages. Integrated directly with the Prometheus alerting pipeline.

Correlate WAL growth, dead tuple spikes, and connection storms automatically. AI surfaces root cause — not just the symptom. Grafana + LLM correlation layer, end-to-end.

AI infra decisions at platform level: latency vs cost trade-offs, embedding model lifecycle in Kubernetes, FinOps for vector workloads, CI/CD for AI pipeline schema changes.