Blog
Technical deep-dives, case studies, and field notes from working with customers on cloud and database engineering.
-
How to Identify the Best RAG Model for Your Use Case
A vendor-neutral decision framework for choosing every model in a RAG stack — embedding, reranker, and generator — by the failure you can't tolerate, not by leaderboard rank.
-
Guard PostgreSQL Lock Remediation: Authority and Evidence (Part 2)
Build a guarded PostgreSQL lock remediation path with fresh PID checks, approval IDs, fixed SQL, cancel versus terminate semantics, and an honest LangGraph baseline verdict.
-
Find the Root Blocker: Build an Evidence-First PostgreSQL Lock Agent (Part 1)
Diagnose PostgreSQL lock issues with a read-only agent: trace the blocking chain with pg_blocking_pids, distinguish deadlocks from lock timeouts, and emit one machine-checkable report.
-
Confidence Follows Evidence: A Seven-Day Reset for Performance Under Pressure
A behavior-first system for rebuilding task-specific confidence through small mastery wins, environmental design, and measurable experiments
-
Quality Is a System, Not a Label: How to Audit AI Training and Evaluation Data
A constructively skeptical eight-control audit for human, synthetic, training, preference, and benchmark data
-
Top AI Agent Evaluation Frameworks to Know in 2026 - Pick by the Layer You Need to Test, Not by Star Count
Compare the top AI agent evaluation frameworks in 2026 by the layer you need to test - trajectory, CI, observability, benchmarks - not by GitHub stars.
-
I Added RAG to My Support Chatbot and It Got Worse — Here’s What Actually Fixed It
A controlled three-system experiment holding the base model constant. Naive RAG scored below the ungrounded baseline — and a tiny QLoRA intent router (0.14% of the parameters, trained on a free Colab T4) fixed it and topped every quality metric.
-
Read-Only Is a Lie: The Postgres MCP Server Mistakes to Avoid When You Wire an Agent to Your Database
readOnlyHint is advice; the database is enforcement. The exploratory-to-operational spectrum — and the four independent layers that earn ‘read-only’ for real when you wire an agent to Postgres.
-
When Not to Use Postgres: A Decision Framework for the Four Walls Where One Engine Isn't Enough
One ACID engine now absorbs vectors, time-series, queues, search, and documents. The consolidation case for defaulting to Postgres — and the four specific walls where I still reach for a specialist.
-
From Prompts to Loop Engineering: The Workflow Shift in AI-Native Development
The four-era arc of AI-native development — prompt, context, harness, loop engineering — and why, in 2026, the unit of work you own has moved all the way up to the iteration loop.
-
How AI Actually Helps You Fix PostgreSQL Performance Problems (and Where It Lies)
AI won’t replace your EXPLAIN ANALYZE instincts — but grounded in real stats, it compresses the diagnose-to-fix loop from an hour to minutes. A DBA’s field guide to where it helps and where it lies.
-
AI Agent Evals: Production Readiness Guide
Benchmarks tell you whether an agent can solve a task. Production evals tell you whether it will behave safely when the task gets messy.
-
Build the Eval System: Three Graders, 38 Tasks, and the $3-8 Safety Net (Part 2 of 2)
The complete practitioner’s guide: three grader types ($0/$0/$$), four task patterns, CI architecture, three real regressions caught, and a 4-week playbook — all for $3-8 per eval run.
-
AI Agent Evals: Why SWE-bench Isn't Enough Before Production (Part 1 of 2)
Your AI agent scores 78% on SWE-bench. It also just told a developer it deployed infrastructure — without calling a single tool. Here’s what benchmarks miss, and the $0 eval that catches it.
-
Spend Fewer Tokens, Get Better Code: A Context Engineering Guide for AI Code Assistants (Part 1 of 2)
Anthropic cut tool context by 85%. Accuracy improved from 49% to 74%. Five context engineering practices that make your AI code assistant produce better output — while spending fewer tokens.
-
Invisible Compound Savings: Caching, Workflow Discipline, and the Habits That Add Up (Part 2 of 3)
90% of your AI prompt context repeats across every request. Prompt caching gives you 90% off. The retry tax costs you 1.4x. Here is how structural habits compound into invisible savings.
-
The 120x Spread: Understanding What You Pay For and When It Matters (Part 3 of 3)
The cheapest AI model costs 0.25x. The most expensive costs 30x. A three-tier task taxonomy for matching model capability to task complexity, plus the complete three-layer optimization playbook.
-
PostgreSQL EXPLAIN BUFFERS: How We Cut Checkout Latency 96%
A real-world e-commerce case study: one word added to EXPLAIN ANALYZE diagnosed a checkout regression from 50ms to 1.2s that three days of network debugging missed.