About and Contact

Updated

The paper

Generative Artificial Intelligence for Small to Medium Size Organizations: A Comprehensive Strategy for On-Premises Deployment is an independent, self-published white paper for technical and business leaders at small and medium organizations in the United States who are deciding whether to run generative AI on their own infrastructure. It covers phased GPU hardware procurement, open-weight model selection, how cloud AI providers compare, infrastructure for AI agents, security architecture, skills and organizational readiness, and the costs and returns that make up the investment case.

The paper is published in versions. The revision history lists what changed in each one.

The author

Emilio Borges is a senior computer engineer and data systems architect with over ten years of experience building large, distributed, multi-threaded, real-time data systems in modern C++ on Enterprise Linux and leading the technical teams that deliver them. That experience includes directing the design, build, deployment, and maintenance of distributed server systems that process and analyze real-time data across multiple machines, applying MLOps principles and CI/CD pipelines, and delivering mission-critical capabilities. As an NVIDIA Certified Associate in AI Infrastructure and Operations, Emilio has the qualifications to design, deploy, and operate on-premises AI capabilities on NVIDIA accelerators. This work covers model selection, serving, and fine-tuning; air-gapped deployment; retrieval-augmented generation pipelines, agent tooling, and prompt and context engineering; the vector, document, relational, and graph data stores those systems depend on; and the capacity planning, observability, data governance, security, and workforce integration that make them dependable in practice.

How the paper is produced

  • Sources. Claims rest on cited sources that pass the source-quality framework. Each reference shows its class: primary, secondary or contextual.
  • Verification. Before a section is published, every citation is checked against its live source, factual errors are corrected, and figures are cross-checked against the other sections. Reference links are rechecked automatically, and each reference records the date it was last accessed.
  • AI assistance. AI assistants are used throughout: to build the research registry, draft and revise sections, review edits against the house style, and convert citations. They also helped build this website. The author works the way a film director does. He sets the scope, the arguments and the standard each section must meet, and AI assistants work as the crew under that direction: researchers building the research registry, writers drafting and revising sections, editors checking edits against the house style, and technicians converting citations and helping build this website. As director, the author reviews every scene and keeps final cut. He checks the sources, edits the work, decides what is published, and is responsible for every claim. AI is one more tool in that work. When it is used responsibly, effectively and ethically by people who know how to use it, it makes work faster and more efficient and helps them produce higher-quality results.
  • Corrections. Confirmed errors are fixed in the next version and noted in the revision history.

Source-quality framework

A source is cited only if it would hold up in the reference list of a peer-reviewed venue. Three tests decide this: accountability (a named author or institution stands behind it), independence (it does not profit from the conclusion it supports), and accuracy (the cited fact appears in the source as reported and is still current). Independence is applied most strictly. A vendor is a sound source for what its own products are, do and cost. A claim that its product outperforms a competitor’s also needs a source that does not profit from the answer.

Each cited source belongs to one of three classes:

  • Primary. Peer-reviewed journal and conference papers, standards, laws and executive orders, government and institutional statistics, benchmark results with published methods such as MLPerf, and vendor specifications, documentation and pricing.
  • Secondary. Industry and analyst research that discloses its method, vendor blogs and announcements cited for product facts, preprints awaiting peer review, legal and policy analysis from established firms and publishers, and first-party measurements with a reproducible method. Where a secondary source’s limits bear on a claim, the text says so.
  • Contextual. News reports, opinion pieces, and practitioner blogs and guides. These sources document events and announcements, record what named people have said, and give readers background. Figures drawn from them are checked against primary or secondary sources where those exist.

Sources that fail the three tests are not cited. These include content-farm and search-optimized articles, undated or anonymous posts, forum threads, aggregators that restate a source the paper can cite directly, marketing pages that claim performance or return on investment, pay-to-publish journals, and any source that cannot be located.

Contact

Write to contact@emilioborges.com about:

  • errors, corrections and questions about the content;
  • permission to reuse material beyond the license;
  • accessibility problems with the website or the PDF edition;
  • privacy questions;
  • speaking invitations and other inquiries.

A clear subject line, such as “Correction: [section title],” helps route your message.

Comments and moderation

Each section ends with a comment thread hosted on GitHub Discussions. Questions, corrections, counterarguments and practical experience are all welcome. To keep the threads useful, comments may be hidden or deleted, and threads locked, when they contain spam or unrelated promotion; personal attacks, harassment, threats or hateful content; confidential or personal information, credentials, or details of unpatched security vulnerabilities; material that infringes someone else’s rights or breaks the law; or discussion unrelated to the section. Moderation is at the author’s discretion, and not every comment will receive a reply. Comments are public, belong to the people who wrote them, and follow GitHub’s Terms of Service and Community Guidelines, under which GitHub may also act on them. To report a comment, use GitHub’s report option or write to the address above.

Contents