Forward deployed AI engineering · Based in Mexico, working remote
AI in production inside your operation, one workflow at a time.
Most companies already pay for ChatGPT, Gemini or Copilot. Few have AI running a process that matters. We work inside your team, pick the one workflow worth automating first, and take it to production with evals, a shadow run and a kill switch. It starts with a 2-week diagnostic at a fixed price.
2 weeks · USD 12,500 fixed · one client at a time
Three steps, each with a written deliverable.
Diagnostic · 2 weeks · USD 12,500 fixed
We map the process with the people who run it: the systems it touches, the data as it really is, and what must never happen. You get a first scope cut to what can reach production, an eval plan with the threshold that defines good enough, and a shadow-rollout plan with a kill switch.
First workflow to production · 4 to 8 weeks
The workflow goes live behind evals, first in shadow next to the people who do it today, then for real. The price is fixed at the end of the diagnostic, before we start. We work by deliverable, with scope and dates agreed up front.
Your team owns it
Runbook, evals, dashboards, and a team that has already run it. Anything your software or model vendor should fix goes back to them in writing, so you are not paying us to maintain workarounds.
Processes with documents, rules and money on every error.
Insurance · medical claims
Pre-authorizations and claim reviews
The pre-authorization packet checked against the policy and the fee schedule before a claims reviewer sees it, and the review drafted with every line traceable. On the hospital side, the mirror image: the packet complete before it is sent.
Manufacturing · Mexico's IMMEX program
Inventory discrepancies before the audit
Customs entries against the Anexo 24 inventory control against physical stock: discrepancies explained and corrections ready before the self-audit or a tax authority visit.
Shippers · Mexican freight documents
Freight reconciled before month-end
Every carrier invoice and its Carta Porte matched against the shipment and the proof of delivery, so errors get fixed or rejected before close instead of costing the tax deduction.
Your process
Something else?
If it runs on documents and rules and a mistake costs money, send it. I will tell you in one reply whether it is worth a diagnostic.
Most firms say “senior engineers”. Here is what ours are graded on.
Every engineer who works on a project with me, and every candidate we send a hiring team, clears the same case. It opens with a discovery call where the model plays your customer and answers only what is asked, up to eight questions. Then a timed deployment writeup, scored 1 to 5 on seven dimensions, with every hidden trap caught or missed.
Discovery and constraint mapping
What they ask before they build anything. The question that decides whether grounding is even buildable usually goes unasked.
Intent and scope surgery
Can they cut a first use case that reaches production in weeks, and say what is out of scope out loud.
Agent architecture
Narrow typed tools, orchestration they own, no mega-tools that hide four failure points behind one call.
Eval plan
Do they catch the failure that matters, or the one that is easy to measure.
Rollout and stakeholder design
Shadow, canary, kill switch, and who signs off. A launch date is not a deadline for the model.
The asks
What they demand from the customer before writing code. The best candidates ask for the thing nobody wants to give.
Product feedback loop
What they push back into the product instead of hacking around it for one customer. The instinct that separates an FDE from a contractor doing bespoke integration.
Your data, your cloud, your off switch.
Built in your environment
Where it matters, we deploy inside your own cloud account (AWS Bedrock, Azure or Google Vertex), so the data does not leave it.
No training on your data
Models are called through business APIs whose terms exclude training on your inputs.
Every model call traceable
Inputs, outputs and the decision taken are logged, so your auditor or compliance team can follow any single case.
A person signs off
Nothing goes live without a shadow run, a threshold you agreed to in writing, and a kill switch your team controls.
Hiring FDEs, or want to be one?
Hiring FDEs
Same bar, for your candidates.
- ✓Screen your own applicants: the first five free, then $49 per graded candidate
- ✓Or placement on contingency: nothing up front, a fee only when you hire
- ✓Every candidate arrives with the seven-dimension review and the thing they built
120 FDE roles are open on our board right now, across 14 companies.
You are an engineer
Get graded, get presented, or train until you clear it.
- ✓Sit the free timed case. One hour, graded on the same seven dimensions
- ✓Clear it and we present you to companies hiring FDEs. You are never charged
- ✓Miss it and train: the 14-day sprint, the coach and the dossiers, $249 once, refunded if we place you
What people ask first.
What does the diagnostic cost, and what do I get?
USD 12,500, fixed, over 2 weeks. Four written deliverables: a map of the process, its systems and what must never happen; a first scope cut to what can reach production; an eval plan with the threshold that defines good enough; and a shadow-rollout plan with a kill switch. If the process is not worth automating yet, the diagnostic says so.
Who does the work?
I do. I lead every engagement and take one client at a time. Engineers who join a project work on it with me, and every one of them has cleared the case published on this page.
Do you work on site?
Remote first, with on-site sessions when the work needs them. We sell a deliverable with agreed scope and dates, not hours or seats.
Which models and platforms do you use?
Whatever fits your stack and your contracts: Claude, OpenAI or Gemini models, through your own cloud account where possible. We resell nothing and take no commission from any vendor.
How much does it cost to hire an FDE through A10X?
Nothing until you hire. Placement is contingency: a percentage of the hire's first-year base salary, payable only when they start, quoted against the specific role before we send anyone. Every placement carries a 90-day replacement window. Screening your own applicants is priced separately: the first five are free, then $49 per graded candidate.
Do you charge engineers?
Never for placement. The company that hires pays the fee, which is also why we do not sell or list candidate data to anyone. The training is a separate, optional $249 purchase.
Armando Gonzalez
Two years building AI systems in production, forward-deployed work included, and the hiring side of FDE loops. Based in Mexico, working in Spanish and English.
I lead every engagement, I wrote the cases and the rubric, and I answer this address myself.
Send me the process. I will tell you if it is worth two weeks.
One email: the process in two lines, the systems it touches, and what one mistake costs. I read and answer it myself.
Book the diagnostic →Hiring FDEs? Send me the role · Engineer? Take the free case