What it covers
Model routing and failover — multi-vendor without multi-refactor
RAG pipelines — ingestion, retrieval, and reranking that measurably work
Evaluation harnesses — golden sets and regression evals before you ship a prompt
Agent architecture — tool design, guardrails, and observability from day one
Cost engineering — per-feature attribution and routing policies that cut spend
What "done" looks like
A pipeline with a scoreboard
Retrieval and generation quality measured against a golden set, with eval runs in CI — so 'the model got worse' is a graph, not a vibe.
A routing layer you own
Vendor switches become config changes. Your prompts, traces, and spend data live in your infrastructure.
How pricing works · USD
- Fixed sprint
- $2.9k–$9.9k fixed, 2–4 weeks
- Fixed scope, fixed price, agreed before work starts. Scope changes are re-quoted, never absorbed.
- Optional retainer
- from $1.4k/mo, disclosed before you commit
- Scoped after a 15-minute call — always disclosed before you commit, never open-ended, never a default.