Remote AI engineer for San Francisco Bay Area companies

I am a Senior AI Engineer and I work remotely from Monterrey, Mexico, with companies in San Francisco and the Bay Area, mostly on evaluation, agents and MCP servers. I am not in California: my working day begins one or two hours before yours, and everything I deliver lands in your repositories and in writing.

San Francisco and the Bay Area are known for software and AI companies, venture capital, biotechnology and financial technology. It is also where many AI model companies are, so nobody there needs a language model explained. This page says what an engineer outside California adds to that, sector by sector, and where the published evidence stops.

Two hours in winter, one in summer

Monterrey stays on one hour all year. San Francisco is two hours behind Monterrey from early November to mid March, and one hour behind from mid March to early November. I am at my desk before your day starts, so a question left in the evening can have a written answer when you arrive. Calls go in your first hours. The rest travels through pull requests and review comments.

Software and AI companies

A team that ships a product on a model usually has the model part solved. What is missing is a way to know whether last week's change made the product worse. That is LLM evaluation: scripted cases, a judge calibrated against human labels, and a gate in the pipeline that can refuse a release. The published example is the evaluation gates of a multi-agent system that serves a higher-education group with 700,000+ students.

The second request is a connection. Either your product has to be reachable by other people's agents, or your agent has to reach a customer's system without being handed the keys to all of it. I build both as MCP servers, the subject of the service called MCP servers and integrations. The open example is an Airtable MCP server that lets an assistant query a database in plain language. For teams that want to see how I reason before they write, my research on agents is pre-registered and public.

Venture firms and the companies they fund

A venture firm receives more decks, data rooms and investor updates than its partners can read. That is retrieval over documents the firm already holds, and it falls under internal tools with AI and RAG and data platforms. What this site can show is an assistant that answers from one fixed corpus, built for a lean manufacturing practice in Monterrey. The subject could not be further from venture capital. The discipline is the same one: the answer comes from the documents.

A newly funded company has a different need: one senior engineer for one bounded piece, before a team exists. That is a contract engagement with its scope in writing.

Biotechnology

Biotechnology companies keep what they know in controlled documents: protocols, study records, regulatory submissions. No case study on this site comes from biotechnology. What transfers is that corpus method, together with access to the system of record that is narrow and logged.

Financial technology

A financial technology company answers to the rules of banking even when it does not look like a bank. An agent that touches an account must leave a trace: what it did, with which data, on whose authority. No case study on this site comes from banking. What transfers is the standard I build to, set out in the guide on agent auditability, and a gateway that stands between an agent and every tool it may call.

Starting, and what you will not get

Email me the role or the problem. I reply in writing with what I would build first, the access it requires and what I will not promise. What you will not get is an engineer in the room. If the work depends on that, someone local is the right choice.

Common questions

Do you live in San Francisco or elsewhere in the Bay Area?

No. I live and work in Monterrey, Mexico, and everything I do for a Bay Area company is done remotely. A search with a city in it usually expects a local engineer, so I prefer that you read this before you write.

How far apart are the clocks of San Francisco and Monterrey?

San Francisco is two hours behind Monterrey from early November to mid March, and one hour behind from mid March to early November. Monterrey keeps the same hour all year, so only the California side moves. Nine in the morning in San Francisco is eleven in Monterrey in winter and ten in summer.

Our product already runs on a model from one of the AI companies nearby. What would you add?

The parts a model does not bring. Those are the tests that say whether your product got better or worse after a change, the narrow and logged connections to your data, and the record of what the agent did. A stronger model makes the first one more urgent, because every upgrade moves behavior you had already accepted.

Can a startup engage you for one defined piece of work?

Yes. I take contract engagements with the scope in writing, and I am also open to a full-time senior role. For a small company I would propose one bounded piece first, such as an evaluation suite for the agent it already has.

Is there a case study from biotechnology or from banking on this site?

No. No case study on this site comes from biotechnology or from banking. What transfers to both is method: answers drawn only from a controlled set of documents, access to systems of record that is narrow and logged, and tests that can block a release.

By city

Open to senior AI engineering roles and contract engagements, remote in the US, Mexico and Latin America.