AI Product Ship-Readiness Audit

Founders and small teams whose AI product has been "almost ready" for months. You've built fast — maybe with AI coding assistants, maybe with contractors, maybe both — and now you have three half-connected subsystems, a demo that impresses, and no clear answer to the question: what exactly stands between this and a paying customer?

If you're planning a rebuild, raising on a launch date, or deciding whether to hire, this audit gives you the ground truth first.

A written report, delivered within five business days, containing:

• A system map of your actual architecture — every component tied to the source file that defines it, not the architecture you remember designing

• Three root blockers, each classified as an architecture, execution, or scope problem, with every claim cited to specific files and lines in your repository — you can check my work

• Your smallest sellable workflow: the shortest path through your existing code that delivers real value to a paying customer, and the precise gaps that remain on it

• A critical path to launch — what to build, what to fix, and just as importantly, what to stop building

• Assumptions and risks, stated plainly

Plus a 60-minute readout call to walk through the findings and your options.

• Intake (30 min call + questionnaire) — what you believe the system does, what's blocking you, what "launched" means to you

• Repo handoff — you provide a snapshot of the codebase

• Audit day — 8 focused hours running a systematic pipeline: system mapping, documentation-vs-code drift analysis, retrieval pipeline review, subsystem coupling, evaluation infrastructure, policy enforcement, and end-to-end workflow tracing

• Report + readout — findings, evidence, and a decision-ready path forward

Your code never leaves one machine. The entire analysis runs on self-hosted models on air-gapped hardware — no cloud APIs, no third-party AI services, no external internet connection during analysis. Your intellectual property is never used to train anything and never transits anyone else's servers. Full methodology disclosed in every report.

Evidence, not vibes. Every finding cites file:line. If I claim your medical-advice boundary exists only in a prompt and not in code, the report shows you where I looked and what I found. You don't have to trust my judgment — you can verify it.

Fixed scope, fixed price. $3,000. No discovery phase that becomes a retainer. One day, one report, one call.

I'm Bruno Gonçalves. I hold a PhD in the Physics of Complex Systems, held a tenured faculty position at Aix-Marseille Université, and completed a Data Science fellowship at NYU's Center for Data Science. I've spent two decades analyzing complex systems — for the last several years, the ones built out of LLMs, retrieval pipelines, and agents. I run Data For Science, where I teach thousands of practitioners how to build and evaluate production AI systems.

How do you handle my code? You share a snapshot; it's analyzed exclusively on isolated, self-hosted infrastructure and deleted after delivery unless you ask otherwise. An NDA is welcome.

Do you use AI in the audit? Yes, transparently: locally hosted models perform systematic scans under my direction, and every AI-assisted finding is verified against the source before it enters the report. The judgment: what matters, what blocks you, what to do, is mine. Each report includes a full tools disclosure.

What stacks do you cover? Python and TypeScript/JavaScript ecosystems primarily, including the common AI product stack: LLM APIs, RAG pipelines, agent frameworks, and messaging integrations.