Fig. 00 Forward Deployed · Applied AI Engineering

AI agents fail on your data, not on the model.

I’m a senior software engineer & architect. I build the systems, the data pipelines and the safeguards that make AI agents valuable in production — not just in the demo.

Book a readiness assessment Read how I work

Systems · Data · Safeguards — the eighty percent nobody sells

Fig. 01 Where this comes from

The demo was never the hard part.

A pharmaceutical client paid a global data provider for account-level distribution data across every retail and non-retail channel in the country. It arrived as portal downloads, shaped for the vendor — not the buyer.

Every month, an analyst rebuilt the same transformations by hand. Territory changes broke the mapping. Historical restatements silently overwrote numbers that had already been reported. Sales compensation ran on the result.

The data was never the problem. The eighty percent nobody sells — ingestion, reconciliation, mapping, validation — was the problem.

Agents are the same shape, one technology cycle later. The interface changed. The hard part didn’t.

Fig. 02 Honest by default

When you shouldn’t hire me.

  • 01

    You want a chatbot on your website.

    Buy one — there are good ones for $30/month.

    Buy off-the-shelf
  • 02

    Your data already lives in one clean system.

    An off-the-shelf agent platform will get you there faster and cheaper.

    Use a platform
  • 03

    You need a demo for a board meeting next week.

    I build things that hold up in production, which takes longer than a demo.

    Ask a demo shop
  • 04

    You haven’t decided what decision the agent needs to make.

    That conversation comes before any engineering does.

    Start with the decision

Fig. 03 What I do

Four services, one discipline.

01 Embedded

Forward Deployed Engineering

I work inside your systems until the thing runs in production — not until the demo works. Embedded, with production access and the authority to fix what breaks.

  • production access
  • embedded, not advisory
  • owns the outcome
02 Scoped

Applied AI Engineering

Agents scoped to one real decision, with guardrails for what they may do alone, what needs approval, and what they must never touch.

  • one real decision
  • guardrails by design
  • approval gates
03 The middle layer

Workflow Automation

The unglamorous middle layer — ingestion, reconciliation, mapping — that makes a purchased data feed or a new tool actually usable.

  • ingestion
  • reconciliation
  • mapping
04 Auditable

Data Science & Data Engineering

Pipelines and models built to be audited, not just demoed: versioned logic, reconciliation reports, lineage back to the source.

  • versioned logic
  • reconciliation reports
  • lineage to source

Fig. 04 How we start

An engagement ladder, not a subscription.

Step 01

Readiness Assessment

2 weeks · fixed price

A systems and data-quality audit, a task-by-task automation feasibility review, and a risk classification of what an agent may decide alone versus what needs a human.

You leave with a ranked plan — including an honest “not yet” list.

Step 02

First Agent in Production

4–6 weeks · fixed scope

One workflow, fully scoped, with a measured baseline so the improvement is provable, not assumed. Guardrails, logging, and approval gates built in from day one.

Improvement you can prove, not assume.

Step 03

Operate & Improve

Ongoing · monthly

Monitoring, evaluation, and iteration as your business changes. The part a product can’t do for you.

Agents that keep up with the business.

Start

Every ladder begins on the ground.

Two weeks, fixed price, no lock-in. If the honest answer is “not yet,” you’ll get that too.

Book the assessment

Fig. 05 Ledger

Reconciled, not just reported.

One engagement, end to end: a pharmaceutical client’s distribution data pipeline, rebuilt — with one item deliberately left open.

✓ 0 hrs/mo Manual rebuild time eliminated — monthly vendor ingestion went from ~30 hours of hand work to zero.
✓ 100% Historical restatements reconciled automatically — prior figures stay accurate and intact.
✓ weeks → hours From vendor delivery to a dashboard the commercial team signs off on.
⚠ 1 open item Territory-realignment edge case kept under intentional manual review. Thoroughness over theater.

Fig. 06 On the record

Tool-agnostic. On the record.

I don’t resell a platform. If an off-the-shelf agent workspace solves your problem, I’ll say so and help you set it up. I get paid to make the right call — not to make you an integration.

  • 01No platform kickbacks or reseller margins
  • 02Recommendations written down, dated, and open to review
  • 03If the answer is “don’t automate this yet,” that’s the answer

Fig. 07 Contact

Start with an honest assessment.

Two weeks, fixed price. You’ll know exactly what’s worth automating before either of us commits to more.

Book a call

Based in Ecuador — working with teams across LATAM and the US