A fixed price only means something if the edge of it is written down.
Three tiers, one entry offer that costs you nothing, and a published list of what each one doesn't include. No hourly billing, no estimate that becomes a change order in week three.
Entry
AI Reliability Audit
Free
five working days
Teams shipping AI features who need an independent check before launch.
- Precision, recall and F1 against a goldset built from your corpus
- Refusal rate on out-of-scope questions
- Citation correctness
- p50 / p95 latency and token cost
- Server-side authorization review
- A written report listing the failures
Failures are published alongside the scores. Credited in full against any build that follows.
Request the auditCore
Scoped Build
Full builds start at $3,500
3–6 weeks · milestone-based
A deadline and a scope, with a real deliverable at the end of it.
- Retrieval and document intelligence: extraction, sentence-aware chunking, embeddings, vector search with an HNSW index, answers with sources attached
- Tool-calling agents as LangGraph state machines — conditional routing, validated tool arguments, bounded loops
- Human handoff that carries the whole conversation, not a summary of it
- A committed test suite and a benchmark run against a goldset before handover
- Token and latency reporting wired in from the first milestone
- Source, deployment, documentation and a handover call
Quoted after a free reliability audit, once the scope is real. No hourly surprises.
Scope a buildPrimary
White-Label Delivery Capacity
From $8,000 per project
or from $12,000 / month · NDA before the first call
Software agencies and product studios with signed AI work and no one to deliver it.
- You keep the client, the relationship and the margin
- I own the build end to end, under your brand, in your repo or mine
- Your name is on the proposal; my name is not in it
- Weekly delivery at your cadence
- Retrieval benchmarked against a goldset you approve
- Source and infrastructure transferred to your accounts
Not a sales call, and not a seat on your client-facing calls.
Request a capacity conversationNot included, published rather than discovered
- Model and cloud usage fees — billed by the provider, not by me
- Third-party licences (vector stores, transcription, telephony, payment providers)
- Ongoing maintenance after the 30-day defect window
- Work outside the agreed scope — quoted as a change order, never absorbed silently
- A platform subscription, a per-seat charge, or any dependency on me after handover
Why it costs what it costs
The variable is not how long I take — it's what breaks if I get it wrong. Retrieval that quietly returns the wrong chunk, a tool call that executes an unvalidated argument, a handoff that arrives without the conversation attached: each one is cheap to build and expensive to discover in production, which is why the measurement work is inside the price rather than bolted on afterwards.
A cheaper quote can still carry a higher total cost, and a higher fee does not prove better results. Compare the same scope, the same acceptance criteria and the same handover. If a tier above doesn't fit, the one below will be quoted honestly rather than padded up to reach it.
This is not for you if…
- You want the cheapest credible option. There are cheaper ones; they will cost you more later.
- You want a quote before anyone looks at your data. Scoping without looking is guessing, and you'd be paying for the guess.
- You want an accuracy figure guaranteed in advance. I will not commit to a number I haven't measured on your corpus.
- You have no deadline and no scope yet. Come back when you do — I'd rather tell you that now than spend your time.
I'd rather lose the sale here than three weeks in.
Start with the audit. It costs nothing and it settles most of it.
Five working days on your own documents, a written report you keep either way, and a fixed price afterwards if you want to go further. If the answer is that you don't need a build, that's what the report says.