I help technical leaders make the architecture, reliability, evaluation, cost and operating-model decisions that separate promising AI prototypes from sustainable production systems.
Send one concrete technical decision, the alternatives you have considered and your deadline.
Focused engagements. Clear deliverables. No long consulting transformation programme required.Built from operating the problems, not presenting slides about them.
Each offer is fixed in scope and designed to produce a decision, a risk map or a clearer next step.
Find the biggest production risks before they become platform debt.
Prioritized findings, a risk map, fix now / next / later / don’t build recommendations, architecture observations, a 90-day action plan and an executive readout.
Make one expensive GenAI architecture decision with better evidence.
A concise decision memo with a recommended option, rejected alternatives and why, major risks and implementation guardrails.
Senior technical judgment without another full-time leadership hire.
A limited number of sessions each month, asynchronous document review and concise written recommendations.
The problem is rarely a lack of AI ideas. It is usually a decision that has not been made clearly enough.
“We have five GenAI pilots and no common production architecture.”
“Every team is selecting its own models and tooling.”
“Our LLM bill is increasing but we cannot explain why.”
“We have observability but cannot answer whether the system is actually getting better.”
“We need to choose between an internal platform and vendor tooling.”
“Our RAG system works in demos but fails unpredictably in production.”
“We are adding agents or MCP and are unsure where authorization should live.”
“We need an independent view before committing significant engineering effort.”
Good GenAI architecture should emerge from actual product and operational requirements. The correct answer is sometimes not to build the platform yet.
Standardize what is expensive to duplicate. Leave everything else flexible until evidence says otherwise.
Define what “good enough” means for quality, latency, reliability, cost and ownership before choosing more infrastructure.
Share the parts that are expensive to duplicate: access, telemetry, evaluation and guardrails. Keep product choices close to the product.
Use explicit trade-offs, measurable tests and decision records. Do not turn an early assumption into a permanent platform.
A system without a clear owner will eventually be called a platform problem, even when the underlying issue is organizational.
The objective is not to maximize consulting days. It is to resolve a high-value technical decision, identify material risks or unblock a team.
30 minutes to establish whether the problem is suitable.
Agree on the question, evidence needed, deliverable and price.
Review architecture, documentation, metrics and relevant systems.
Receive concise findings and actionable recommendations.
Diogo Santos is a technical leader with a background spanning computational biology, data science, machine learning and GenAI systems.
He has led multidisciplinary technical teams and worked on the transition from experimentation to production across ML and GenAI.
His current interests include production GenAI architecture, LLM platforms, agent systems, evaluation, reliability and the operating models required to make AI useful at organizational scale.
Published technical work on agent boundaries and production architecture.
More writing on Medium ↗
Short answers to the questions that usually matter before a first conversation.
The engagements primarily focus on architecture, technical judgment and decision support. Implementation support can be discussed when it materially improves the outcome, but the service is not positioned as outsourced development.
Yes. The typical model is to work with the existing technical leadership and engineers rather than replace them.
No. The work can be useful for both scale-ups and larger organizations when the technical decision is important enough to justify independent senior review.
Not as the core service. The focus is technical and operational decisions where architecture, production constraints and engineering trade-offs matter.
Yes. In fact, that is the preferred starting point.
A short independent review can be much cheaper than six months of platform rework.
In your email, include:
diogofcul@gmail.com
Your details are used to answer this enquiry. Privacy details.
Diogo Santos receives the name, work email, company, optional role and problem you provide, to assess and respond to your advisory enquiry. Enquiries are normally removed after 180 days through routine site use; if an engagement begins, relevant records may be kept longer. The legal basis is taking steps at your request before a possible engagement. OpenAI Sites and Cloudflare process the data to host the site and save the enquiry. Resend processes these details only to deliver the notification email to Diogo. You can ask to access, correct or erase your details by emailing diogofcul@gmail.com. You may also contact your local data protection authority. Site events are counted by day without cookies, IP addresses or visitor identifiers.