Skip to content
TechFabric

AI development · San Francisco, CA

AI development for San Francisco companies

The Bay Area needs no description of its software industry. What is less obvious is how much of the AI work here is now maintenance: systems built quickly in the last two years that have to be made durable, governed and affordable.

The demo is not the problem here. The problem is an agent that loses its context on deploy, a retrieval layer nobody can audit, and a model bill nobody forecast. That is the half we build.

Founded

2017, and still independent

Bench

115+ people, 80 Databricks-certified

Experience

Fifteen years average, per engineer

Serving

San Francisco, Oakland, San Jose, Palo Alto

What ai development covers

Agents that hold state

A task that runs for three days, calls six systems and waits twice on a person cannot live in a request handler. We build agents on a durable runtime, so a deploy in the middle of a run does not lose the run, and a retry does not send the same email twice.

Retrieval that cites what it used

An answer nobody can trace is an answer nobody can defend. Retrieval is built so every response carries the documents behind it, which is what makes the difference between a demo and something a regulated team is allowed to ship.

Evaluation, before and after

Without a score, every prompt change is a matter of opinion. We put deterministic scorers and model-based judges in front of the thing so a change ships with a number attached rather than with a feeling.

Human approval where it belongs

Some actions should stop and wait for a person. That gate is part of the system rather than a policy document, so an agent that wants to issue a refund pauses, the approval is recorded, and the run picks up where it left off.

Cost you can see

Model spend, retrieval infrastructure and the review time a system still needs are the numbers that decide whether it stays live after the first quarter. They get instrumented at build time, not discovered on an invoice.

Governed data access

An agent reads through the same grants a person would. On Databricks that is Unity Catalog, so a model that tries to reach a table it has no grant for fails the way any principal fails, rather than because the prompt asked it nicely.

Work a San Francisco buyer would recognise

The AI system we built, run in production, and gave away under Apache-2.0. For a technical buyer here, what we operate is a more useful answer than what we have advised on.

AI systems · Our own product

Fabric

Fabric is a context store, memory and agent runtime built on Databricks, and it is the clearest evidence of what we build: across our clientele, work that took a team of ten now takes three.

The bill arrives before the governance does

Systems built quickly in the last two years tend to reach a model spend somebody is now questioning, and an audit requirement nobody designed for. Instrumenting cost and evidence at build time is what stops the second version being a rewrite.

None of that is unusual for San Francisco, and none of it is a reason to build slowly. It is a reason to decide where the evidence comes from before the first handler is written, because the alternative is reconstructing it later from logs that were not designed to answer the question.

How an engagement runs

  1. 01

    Two weeks, fixed fee

    An AI Readiness Assessment reads what you already run, names the use cases that will survive your data and your permissions, and puts a number on what each will cost to operate. It ends in a written document that is yours either way.

  2. 02

    One system, named

    We pick the first thing to build with you and say what done looks like before starting. A first system that reaches production beats three that reach a demo.

  3. 03

    Build it with your team

    Your engineers are in the repository with ours from the first week. The pattern has to be one they can extend after we leave, which does not happen if they meet it at handover.

  4. 04

    Run it, then hand it over

    We stay on the pager while it settles, because the design decisions look different when the person making them is the one being woken. Handover happens when it is boring, not when the invoice is due.

Where we work from

TechFabric is headquartered in Gilbert, Arizona, in the Phoenix metropolitan area, and works with companies across the United States. We do not keep an office in San Francisco and are not going to pretend otherwise on a page you found by searching for one.

Arizona does not observe daylight saving, so from March to November we are on the same clock as you and from November to March we are an hour ahead. Either way the working day overlaps almost completely.

Most delivery is remote or hybrid because that suits the work. We travel for the parts that go better in a room, which is usually discovery, workshops and the opening weeks of a build.

Headquarters

1530 E Williams Field Rd, Ste 201
Gilbert, AZ 85295

Phone

+1 (480) 681-6806

Also

Amsterdam, Dnipro, Hyderabad

FAQ

AI development in San Francisco: common questions

Do you work with San Francisco companies?

Yes. TechFabric is headquartered at 1530 E Williams Field Rd, Ste 201, Gilbert, AZ 85295, and works with companies across the United States. Arizona does not observe daylight saving, so from March to November we are on the same clock as you and from November to March we are an hour ahead. Either way the working day overlaps almost completely.

Do you have an office in San Francisco?

No. We have one office and it is in Gilbert, Arizona. Most of this work is remote or hybrid because that is what suits it, and we travel for the parts that go better in a room, usually discovery, workshops and the opening weeks of a build.

We built this fast and now the model bill is a problem.

That is the most common brief we get here. Model spend, retrieval infrastructure and the human review a system still needs get instrumented, and then the architecture questions become answerable. Usually the fix is durability and caching rather than a different model.

What kind of AI development do you do?

Agents that carry state across a long task, retrieval that cites what it used, workflow automation that survives a restart, and the evaluation that tells you whether any of it got better. The common thread is that it has to run unattended and be explainable afterwards, which is a different problem from getting a good answer once.

How does an AI engagement start?

Usually with something short and fixed so the cost is known before you commit. The AI Readiness Assessment runs two weeks for a fixed fee and ends with a written deliverable that is yours whether or not you continue.