Technology Consulting & Advisory

What's included

• Architecture reviews: an outside, senior read on what's solid and what's a liability
• Build vs. buy analysis for AI and infrastructure decisions
• AI readiness assessments: what your data, systems, and team need before an AI initiative can ship
• Vendor and infrastructure cost audits, including the kind that found tens of thousands per month in waste

How we approach the work

1. 01

   Establish the current state

   Review the architecture, code paths, operating costs, and team constraints that actually shape the decision. A recommendation without that baseline is guesswork.

2. 02

   Trace the expensive or risky path

   Follow the behavior through logs, infrastructure, and code until the source of cost or fragility is clear. In one RPC audit, that meant finding a bug behind tens of thousands per month in excess calls.

3. 03

   Compare options and hand over a plan

   Set out the tradeoffs, expected effort, and order of work so leadership can make a build, buy, or repair decision and the engineering team can act on it.

Who this is for

• Companies about to commit budget to a build and want a second opinion first
• Teams inheriting a system they didn't build and need to understand fast
• Leadership evaluating whether an AI initiative is ready to be funded

Related work

• Cutting tens of thousands per month in excess RPC spend — An infrastructure and code audit for a crypto company that uncovered a bug driving tens of thousands of dollars per month in unnecessary blockchain RPC costs.

Related engineering notes

• Build vs buy: custom AI or off-the-shelf? — Build vs buy for AI: when an off-the-shelf tool wins, when a custom build wins, and the total-cost questions teams forget to ask.
• What it actually costs to do local AI development in 2026: GPUs, models, and the math against Claude and Codex — Local AI cost in 2026, priced out honestly: GPU options and VRAM limits, electricity, and the break-even math against Claude, Codex, and API spend.
• llama.cpp vs Ollama: which should you run local models with? — llama.cpp vs Ollama, decided: how the two relate, where each wins on ease, control, and Hugging Face model access, and why we lean llama.cpp for real work.
• RAG vs fine-tuning: which one do you actually need? — RAG vs fine-tuning, decided: what each one actually solves, when to use which, and why most teams reach for fine-tuning when they should not.