Services

I offer a written second opinion on AI quotes and projects, a paid production readiness review, an agent pilot, MCP integration, implementation of the agreed changes, part-time technical leadership, and code handover.

The prices shown are fixed for the scope described. Implementation, technical leadership and handover are priced after an assessment.

Email first; a 20-minute call if useful.

Where do I start?

The pilot builds the agent and its workflow, with the tools it needs. MCP integration connects one of your systems to an assistant you already use. You pick one or the other, not both in the same engagement.

For a single quote, the second opinion is enough; part-time technical leadership is for ongoing decisions.

Written second opinion

Do you have an AI quote or project, including one for an AI agent or an automation? I read the code or the quote and tell you in writing whether to go ahead, fix it first or stop. I do not run the system.

  • One project or one quote, with 3–5 questions agreed in writing
  • Typical questions about an agent: what it can do without approval, what happens when a step fails, how quality is measured
  • A private opinion of 2–4 pages, delivered within 5 working days of receiving the material and your written confirmation of the engagement
  • The questions to ask before you decide, including on cost
Pricing
€1,562 in total€1,500 + 4% INPS contribution + €2 stamp duty; VAT not applicable (flat-rate regime)

If you then choose the full review (where I run the system and measure it) of the same project, the €1,500 fee is deducted from the review fee (€3,500). So the review costs you €2,082 in total. The credit applies if you confirm the review in writing within 6 months of receiving the opinion, for the same project and the same company. It can be used once, only on the review, and is not refundable as cash.

Get a written second opinion

Production readiness review

Duration
Up to two weeks, from when I receive access and material and you confirm the engagement in writing
Pricing
€3,642 in total€3,500 + 4% INPS contribution + €2 stamp duty; VAT not applicable (flat-rate regime)

I review one working LLM-backed prototype in your repository. I look at where state is lost, how it recovers from a failure, how output quality is measured, and what each request costs. For an agent, output quality also covers whether it picks the right tool with the right arguments and, if it uses RAG, whether it retrieves the right sources. The review does not include code changes.

Starting point

One working LLM-backed prototype in your repository.

Before the review

I agree in writing which repository and workflow are in scope, what access and test data are authorized, and which runs and traces are available. If a measurement cannot be reproduced with the available evidence, I report it as unknown rather than claim a result. I work on your model provider account; you pay for model usage.

What this review is not

It is a four-pillar engineering assessment, not a security audit, penetration test, legal compliance certification, load test, or code implementation. I identify material issues outside scope without claiming to certify those areas.

Deliverables

  • A written assessment of each pillar (state, recovery, output quality, and cost), citing files and runs in the repository
  • A prioritised list of changes with an effort estimate for each
  • A short architecture note of the system as found
Good fit
A team with one working LLM-backed prototype that needs a written production assessment before implementation. The prototype can also be an agent that uses tools, MCP servers or RAG.
Not a fit
A request for code changes in the same engagement, or a system that is still only a slide deck.
Project code, architecture, and resultsSample review (in Italian, with an English summary)

Agent pilot

I build an agent for one workflow in your repository. Irreversible actions wait for an approval, every step is logged, and quality is measured with evals you can re-run. At the end you get, in writing, what the agent may do and what it may not.

Starting point

One defined process to hand to an agent, with a person who approves actions.

Duration
Up to four weeks in total, from when I receive access and materials and you confirm the engagement and scope in writing. Phase 1, the design (up to two weeks), is included.
Pricing
€6,762 in total · model usage not included€6,500 + 4% INPS contribution + €2 stamp duty; VAT not applicable (flat-rate regime)

You can start with the design only (€2,602 in total). If you then choose the pilot for the same workflow, the €2,500 fee is deducted from the pilot fee (€6,500). So the pilot costs you €4,162 in total. The credit applies if you confirm the pilot in writing within 6 months of receiving the design, for the same workflow and the same company. It can be used once, only on the pilot, and is not refundable as cash.

Before I start: what I need from you

One defined workflow and a person on your team who approves actions. Access to your repository, the systems involved and test data I am allowed to use. An account with the model provider, in your name: you pay for model usage.

Deliverables

  • One workflow, with tools and data agreed in writing at the start
  • Phase 1: the workflow design document, with steps, state, recovery, approvals and the eval plan
  • Durable state, resuming from the last completed step after a failure
  • Evals on cases agreed with you, with the cost per run measured
  • The written limits: permissions, approvals, cases where the agent failed in the evals
  • A short operating note: how to run it, read the log and re-run the evals
  • The code is yours under our agreement
  • I work on your model provider account; you pay for model usage.
Good fit
A team or SME with one process to hand to an agent, able to name who approves actions and to give access to the systems and test data. It suits a first agent, before any wider rollout.
Not a fit
Production rollout and ongoing operation, several workflows in one engagement, compliance certification.

MCP integration

I connect your ERP, CRM or document archive to an agent or an AI assistant through an MCP server (Model Context Protocol, an open protocol for connecting AI assistants to other software). The integration starts read-only. Actions, if you want them, come later: each one waits for an approval and is recorded in an audit log.

Starting point

One of your systems with a documented API and a test environment or account.

Duration
Up to two weeks per phase, from when I receive access and material and you confirm the engagement in writing
Pricing
Phase 1, read-only: €3,122 in total€3,000 + 4% INPS contribution + €2 stamp duty; VAT not applicable (flat-rate regime)
Phases 1 and 2, with approved actions: €5,202 in total€5,000 + 4% INPS contribution + €2 stamp duty; VAT not applicable (flat-rate regime)

If you add Phase 2 later: €2,080 in total (€2,000 + 4% INPS contribution, I cover the stamp duty; VAT not applicable).

Before I start: what I need from you

A system with a documented API and a test environment or account. A dedicated account for the server, with no write permissions in Phase 1. An AI assistant that supports MCP, named in the written scope. The server runs on your systems and accounts.

Deliverables

  • Phase 1: one system, with a dedicated account that cannot write
  • Phase 2: each action shows what it will change and runs only after an approval
  • Audit log: who asked for what, who approved, what changed
  • Tests and documentation; the code is yours under our agreement
  • I work on your model provider account; you pay for model usage. The server runs on your systems and accounts.
Good fit
A company that wants an agent or an AI assistant to work with one of its systems and wants to start without giving it write access.
Not a fit
Several systems in one engagement, actions without approval, ongoing operation of the server.

LLM systems from prototype to production

After the production readiness review, or after a pilot I built, I implement the agreed changes in your repository: durable state, deterministic recovery, reproducible evals, and cost under control.

Starting point

The production readiness review, or a pilot I built, and the agreed implementation scope.

Duration
Implementation with an agreed scope, after the review or after a pilot I built
Pricing
From the estimate of the review or of a pilot I built, after a written scope check

Deliverables

  • Production changes committed to the client repository
  • Tests and notes the client team can use on its own
  • A written list of what changed and how to run it
Good fit
A team that has completed the production readiness review, or a pilot with me, and wants the agreed changes implemented.
Not a fit
Standalone demos or compliance certification.
Project code, architecture, and results

Fractional CTO, part-time

I provide part-time technical leadership for teams working with internal developers or external vendors: architecture decisions, code and proposal reviews, vendor coordination, and a delivery roadmap. My experience includes leading technical teams as a co-founder CTO and, under contract, designing a system from requirements through technical delivery.

Starting point

An agreed technical assessment of your team, architecture, and delivery priorities. A working LLM prototype is not required.

Duration
Part-time, ongoing
Pricing
Scope and fee agreed after a technical assessment

Deliverables

  • Stack and architecture decisions written down
  • Quote and code review for vendors or the internal team
  • A roadmap the client can run without me in the room
Good fit
A team that needs an experienced technical lead to make and document decisions on a part-time basis.
Not a fit
A full-time executive role or an engagement requiring a dedicated delivery team.

Code handover

I help your team take over an existing codebase. I deliver the agreed repository, tests and operating documentation, transfer access I am authorized to transfer, and record any remaining dependencies.

Starting point

A scoped review of the existing repository and operational dependencies. It can be a standalone engagement or follow implementation.

Duration
Scope and duration agreed after the repository review
Pricing
Scope and fee agreed after a repository review

Deliverables

  • Agreed repository and documentation handed over, with authorized access transferred
  • Tests and runbooks for the agreed operational workflows, reviewed in a handover session
  • A written list of remaining dependencies on people and external services
Good fit
A working codebase that your team needs to learn to run and maintain.
Not a fit
Long-term outsourced operation or a request to supply additional developers to your team.
Project code and technical documentation

For businesses and professionals with a VAT number (or equivalent business tax ID), for their business activity only.

From LLM prototype to production

If you have a supplier's quote, the first step is the written second opinion. For a working LLM prototype, the first step is the production readiness review; if I built the pilot, implementation starts directly from it. Implementation is a separate, optional engagement. Part-time technical leadership and code handover can start independently, without this review.

  1. 01

    Paid review

    The production readiness review (€3,642 in total) is a written assessment of one working LLM prototype.

  2. 02

    Implementation

    After the review, or after a pilot I built, I implement the changes within a scope agreed in writing.

Need technical leadership or code handover instead? I start with a scoped assessment of the team or repository, not the LLM review.

I publish negative results too, for example the emotional-memory benchmark data.

Words I use

Eval
a set of test cases, agreed with you, re-run with one command to measure whether the system's answers and actions are right.
Audit log
the record of every step, tool call and approval: who asked for what, who approved, what changed.
Recovery
what the system does when a step fails: it resumes from the last completed step, stops, or asks a person.
Trace
the record of one run, step by step, with inputs, outputs, timing and cost.
Workflow
the sequence of steps an agent follows in a process, for example from a customer request to the reply.

Frequently asked questions

Project delivery

Is this a good fit for my project?
I work with technical teams and founders taking LLM systems from a working prototype to production. I also work with owners and managers of small and mid-sized companies who have an AI quote to check or a process they want an agent to handle. I also offer technical leadership as a standalone engagement, and handover as a separate engagement. Standalone demos and engagements requiring a dedicated delivery team fall outside these services. A typical fit looks like the problem behind orka, which is public on GitHub: requests arriving from Telegram, Discord, Slack, WhatsApp, and HTTP with no single durable queue, routed through one prioritized queue into tracked LLM workflows.
Who owns the code?
The code I develop for your project belongs to you under our agreement. I hand over the repository, documentation, and required access. Third-party libraries and services retain their own licenses and terms; I document those dependencies.
How are timeline and costs estimated?
For LLM implementation, the production readiness review produces the written estimate; after a pilot I built, I estimate from the pilot. For technical leadership I agree on scope, timeline, and fee after assessing the team. For handover I agree on scope, timeline, and fee after assessing the repository. Delivery follows the agreed scope in verifiable iterations.
What happens after delivery?
Maintenance, monitoring and improvements require a separate written scope. A delivery or handover does not include continuous operations or 24/7 on-call coverage.
How do I pay?
By bank transfer. If the fee is €2,500 or less, I ask for full payment at the start. Above that, half at the start and half on delivery, phase by phase. Each payment is due within 15 days of my request, and you get an invoice for each payment. The timeline starts when I receive the material or access and you confirm the engagement in writing, not from payment.
What is deducted from what?
Two things are deducted. If after the written second opinion you choose the full review on the same project, the €1,500 fee is deducted from the review fee (€3,500). So the review costs you €2,082 in total. The credit applies if you confirm the review in writing within 6 months of receiving the opinion, for the same project and the same company. It can be used once, only on the review, and is not refundable as cash. If after the design only you choose the pilot for the same workflow, the €2,500 fee is deducted from the pilot fee (€6,500). So the pilot costs you €4,162 in total. The credit applies if you confirm the pilot in writing within 6 months of receiving the design, for the same workflow and the same company. It can be used once, only on the pilot, and is not refundable as cash. For the MCP integration, if you start with Phase 1 only and add Phase 2 later, Phase 2 costs €2,080 in total, so the total stays €5,202, the same as both phases together.
How do I start?
Write to me by email at info@gianlucamazza.it. Email first; a 20-minute call if useful. If you have a supplier's quote, the first step is the written second opinion. If you already have a working prototype, the first step is the production readiness review. For technical leadership I agree on a team assessment; a working LLM prototype is not required. For handover I agree on a repository assessment.

Scope, systems, and evidence

What is it like to work with a single external technical contact?
I work with you from requirements to production: I make architecture decisions, write and review code, and document operation and recovery. I agree on responsibilities and coordination with your team or vendors at the start.
When do you add AI to a system, and when is it worth it?
Only where it brings verifiable value, with human oversight and costs kept under control. I add AI to a workflow when the output can be checked and the cost is predictable, not as a default. When I build an agent, model output does not act on other systems until it passes a check: a rule, an automated test or a person. In openfatture an invoice waits for a human check before it goes out; reasoning-kernel, a research project, is designed to reduce the risk that untrusted text triggers an action.
Write by email

Written decisions in the public audit