Edgar Martinez

you

Hi! What do you do?

edgar · agent

readprofile.md17+ years · Cogent IBS since 2022

render<Intro />

Hi, I'm Edgar. I design systems on SAP BTP and take them to production, hands-on until they go live. 17+ years, remote from Mexico, on US and EU hours.

<Intro />rendering

SAP BTP Solution Architect

17+ years15+ systems liveMexicoRemote, US and EU hours

15+ systems live since 2008. I stay close to the code on every one: it is how I check that the architecture became what it was meant to be, and on a small project it is how it gets built. CAP, Fiori, SAP Build, integration, and since 2025 machine-learning models running in production.

Message me on LinkedIn ↗Résumé (PDF)

follow upBook a call

you

What kind of problems do people bring you?

edgar · agent

grep-r "as the client described it" cases/6 matches · 2008 to 2024

render<ProblemDeck cases={6} />

Here are six, in the words the client used when they brought each one to me. What I did, what came of it, and what my part was.

<ProblemDeck order="as told" />rendering

01 / 06offline-first SAPUI5

We have tried three times to get this app into production.

2015 · Gentera, microfinance, Mexico

What I did: An offline-first SAPUI5 app for credit origination in the field. It worked with no signal and synced when the agent was back in coverage.
Result: It reached production on the fourth attempt, ours, and 14,000+ field agents used it.
My part: Sr. Mobile Platform Consultant at Nubaj: the offline-sync design, written up later on my blog.

02 / 06machine-learning models · in production

Converting one catalog takes us six weeks.

2025 · a global automaker, spare-parts catalogs

What I did: Machine-learning models (Keras, EasyOCR, Amazon Textract) read callouts and part positions in the drawings; post-processing converts each catalog from the European standard to the American one.
Result: A catalog now takes one week. Delivered on the client's fixed date after an unplanned late start: six complete subsystems (ingestion, recognition, post-processing, review, export and integration with the catalog system), replanned and landed in under three months. Live since 2025 and still being extended.
My part: Architect, the client's main contact and project manager, close to the code throughout.

03 / 06SAP CAP · Build Work Zone

A supplier portal means SAP Ariba, and Ariba is too expensive for us.

2024 · a retail chain in Mexico

What I did: A supplier portal on SAP CAP and SAP Build Work Zone, with Cloud Connector to the SAP backend and Identity Authentication.
Result: 500K invoices a month from 1,000+ suppliers, at up to a fifth of what Ariba would have cost to implement and run. Delivered and in production. Figures from the client case Cogent IBS presented at ASUG Chicago, 4 Nov 2024.
My part: Architecture and delivery lead at Cogent IBS.

04 / 06Business Rules · workflows

Approvals take 20 days, and only IT can change the rules.

2019 · Delivery Hero, Berlin

What I did: Business Rules on SAP Cloud Platform, so the functional teams own approval steps, assignment and thresholds without a ticket to IT; then the project-approval and CAPEX workflows.
Result: Approvals went from 20+ days to 2. That figure belongs to the Finance Systems team, which won the SAP Innovation Award 2019 ↗ (Process Innovator). I joined in April 2019 and built the rules and the two workflows on top of the framework the team had already shipped.
My part: SAP Fiori and Cloud Platform Developer, directly contracted by Delivery Hero.

05 / 06Node.js over RFC

Retailers keep complaining the portal is slow. And it works, so we do not want a big project.

2017 · a US window manufacturer

What I did: The functionality already existed in SAP, as function modules. We tuned them where needed and exposed them over RFC to Node.js, brand new at the time, with React screens on top. No OData overhead, so it flew, and it worked on a phone.
Result: Retailers who had dropped the old portal came back, this time from their phones. Live since 2017, still in use in 2026.
My part: Technical Team Lead at Cogent IBS.

06 / 06OCR · Oracle UCM

The platform has no documentation.

2008 · IMPI, Mexico's industrial property office

What I did: IMPI had to put 80M+ records online; by hand it was impossible, and the migration had a processing-time requirement. Our OCR pipeline automated it, but it had to integrate with Oracle UCM, whose web services came with no documentation. I dug until I found the one page on the internet that explained how to switch them on: an HTML table, saved in the right place.
Result: The services ran, and the pipeline indexed 80M+ records for public access, on time. A team result.
My part: Developer at Qualitas, in my first job.

follow up

you

If we wanted to bring you in, how would that work?

edgar · agent

readoffers.md2 offers · no public price

render<Offers />

Three ways: a scoped project, an architect on your team (fractional, by the hour or by the month, directly or through a consultancy), or a full-time remote role.

For projects there are two offers, both scoped before they are priced. Each starts with a 25-minute call, and fees are quoted per scope once we are on the same page.

<Offers count={2} />rendering

01

Build

A scoped project on SAP BTP.

Custom development with CAP and Fiori, automation with SAP Build, Work Zone, Identity Authentication, Cloud Connector, and integration with S/4HANA and ECC. I design it, lead the team and write code. I have worked across the whole SAP stack, ABAP and PI included: so when a problem lands on the core side, we do not have to stop and look for another team: I can coordinate it, guide whoever is on it, and jump in myself when that is the faster way. You get a working system in your landscape, with tests and documentation. When the build is done, we agree on a transition that fits the project: maybe a second phase, maybe a handover with technical KT, maybe a few hours a month while your team settles in. Whatever shape it takes, your team is not left alone with the system.

Proof: The supplier portal, the catalog solution, version 1 of the CI Accelerator on SAP Store ↗, Business Rules and workflows at Delivery Hero.
Starts with: The system, what it has to do, and by when. I come back with the questions that decide the design, then a first approach.

02

Agent assessment

Two weeks is typical; complex agentic systems take longer. For an agent you are working on now.

You already have an agent, on Joule or anywhere else. I review it as an architect. You get a written report: what works, what fails on your real cases, and the strategy I would follow to get it to production, step by step.

scopeaccessbusiness logictest cases, present and missingcost per run

Needs: A call with the owner and one functional expert, then read access to the agent, a sample of its runs, and the golden test set if one exists, under NDA, after we agree the scope.
Based on: 17+ years of taking systems to production, and the agent work under AI on SAP: a news-verification agent in the works, an open-source agent and talk CT1106 at SAP TechEd 2026.

Are we a fit?

The quickest way to tell: if nothing gets designed, built or modernized, we are probably not a match. That side covers keeping systems running (Basis, upgrades), configuring standard modules, and licenses. If something does need to be designed, built or modernized, that is my side: from understanding the process and what you want to achieve, together with your team, to a system in production.

follow upBook 25 minutes

you

And where did all this start?

edgar · agent

readtrack-record.json8 periods · 2008 to 2026

render<Timeline from={2008} to={2026} />

In 2008, with an OCR pipeline for a public office. Since then, 15+ systems in production. Open a period for my part, what shipped and the proof. The number on the right is the one I can defend; where it belongs to a team, the label says so.

<Timeline periods={8} />rendering

2008 to 2009OCR for Mexico's industrial property office80M+records indexed · team result

My part: Developer at Qualitas, for IMPI. My first job.
Shipped: An OCR pipeline that indexed 80M+ records for public access, integrated with Oracle UCM.

2010 to 2014SAP integration, ABAP and a company-wide AP flow4,000+AP users · team result

My part: Software architect and team lead, then SAP PI consultant, at Indra.
Shipped: For Grupo Bimbo: a web document viewer and approval tool, a robot that distributed invoices, and the accounts-payable flow for the whole company, 4,000+ users in 3 regions, with licensing cost about 80% below the tool it replaced. For Taco Holding: POS-to-SAP integration on 180 terminals. ABAP in WM, MM, FI, SD and HR; import interfaces for Chedraui; an enterprise-link re-architecture for Santander.

2014 to 2016PI interface programs and offline mobile14,000+field agents on the app

My part: SAP PI consultant at Neoris, then Sr. Mobile Platform Consultant at Nubaj.
Shipped: For CEMEX: multiple SAP PI interfaces and a program of mass interface changes, all in production. For Gentera: the offline-first SAPUI5 credit-origination app, live after three earlier attempts had not reached production.
Proof: The offline-sync pattern is written up on my old blog ↗ (PouchDB posts, 2016).

2017 to 2019Node.js on SAP, before CAPLive since 2017RFC + Node.js

My part: Technical Team Lead at Cogent IBS.
Shipped: A dealer portal for a US window manufacturer: existing SAP function modules exposed over RFC to Node.js with React screens, usable on a phone, still in use; a finance-planning suite on S/4HANA and SAPUI5 that passed SAP certification and was published on SAP Store (since retired).
Proof: A recommendation from Sainath Allumalla, on my team at Cogent IBS at the time, quoted under References below.

2019 to 2020Business Rules and CAPEX workflows at Delivery Hero20+ days to 2approvals · team result

My part: SAP Fiori and Cloud Platform Developer at Delivery Hero, Berlin, contracted directly: Business Rules, then the project-approval and CAPEX workflows, all in production.
Proof: Team award: SAP Innovation Awards 2019 ↗, Process Innovator.

2020 to 2023Four contracts: identity, CAP and products other companies sellv1on SAP Store

My part: Contracts, one after the other: Fiori and SAP Cloud Platform specialist at Deloitte (Nov 2020 to Jun 2021); Senior AWS Full-Stack Engineer at Ivaldi Group (Jun 2021 to Mar 2022); Senior Cloud Consultant at SPOSEA (Mar to Jun 2022); architect and developer for Code10, direct (2023).
Shipped: Supplier-facing procurement apps with SAP Work Zone, principal propagation and the program's first mobile approvals app, for a Latin American telecom (Deloitte). Logistics on event-driven AWS, manual operations about 60% lower (Ivaldi). BrightPrice Build extended on CAP and HDI, task time about 25% lower (SPOSEA). CI Accelerator version 1 (Code10). All in production.
Proof: principalpropagationsapcf ↗ · BrightPrice Build ↗ · CI Accelerator ↗

2022 to 2026Architecture and delivery lead at Cogent IBS6 weeks to 1catalog conversion

My part: Architect, client contact and project manager, close to the code throughout. Distributed team across India, the US, Mexico and Argentina.
Shipped: The spare-parts catalog solution for a global automaker, in production since 2025; the supplier portal for a retail chain in Mexico; a solution for a language university.
Proof: A recommendation from Stephanie Wu, Project Manager on the automotive program, quoted under References below.

2026Agents on SAPIn progresssee AI on SAP

What exists: A news-verification pro-code agent, in the works, the zero-token-joule-agents repo, and talk CT1106 at SAP TechEd Berlin.

follow upDownload the résumé

you

Everyone says AI these days. What have you actually shipped?

edgar · agent

readai-on-sap.md1 ML system live · agents in progress

checkproduction-criteria --scope agentsin progress

render<AIStatus />

Machine-learning models in production since 2025. Agents on SAP in progress. I keep the two apart, because the gap between a working demo and a system someone signs off on is most of the work.

<AIStatus honest />rendering

In production

What: The spare-parts catalog solution for a global automaker: the six-weeks-to-one case under Problems, above.
Since: 2025. Still in production, still being extended.
One decision: The model kept reading the codes at the foot of each sheet as callouts. The team wanted to fine-tune it. I had the strip cropped out before recognition, and the errors went away.

decision · catalog solution+1 −1

−fine-tune the model on the codes at the foot of each sheet

+crop that strip out before recognition

the errors went away

In progress

Building: A news-verification pro-code agent with a media company in Mexico: an editor approves every result, and a governance screen logs each decision and the rules frozen so far.

Production means

Real data in the client's landscape
A person reviews the output
A golden test set, so every change proves itself
Evals and tracing on every batch in production
Written sign-off by a named person
At least one real batch processed

What few people tell you: an agent needs a different kind of maintenance than a normal system. Models change even under the same name, so a swap is a project, not a dropdown. That is why I measure a baseline, run a pilot, and only then call the ROI real.

how-i-build.md

How I build now. AI is not only what I ship; it also changed how I build, AI systems and plain deterministic ones alike. I work with coding agents professionally: Claude Code and Codex write most of the code and the first version of the tests. I write the spec, hold the architecture and check the behavior. One session per task, a navigable agents.md, unit tests on every change, and the loop gets better with every project. The code gets written. Taking it to production is still my job.

Models, tools and techniques change almost by the day now. What has worked for me is building and adopting mental frameworks, and knowing how AI works underneath: with that base, when a technique does not exist yet you can build it, and when it does, you know exactly what you are looking for. A good part of my own base came from people sharing what they learned, so CT1106 is me passing it on.

Talk

CT1106 · Zero token architecture for Joule Agents

SAP TechEd Berlin · Tue 27 Oct 2026 · 14:00 CET · 20 min

Let the agent work out the first cases. What it validated becomes a playbook or code that a named person signs, and from then on that case runs without a model.

Talk notes

Open source

zero-token-joule-agents

SAP CAP + LangGraph.js · Node 22+ · MIT

An agent that stops calling the model for the cases it has already learned, and measures how much of its traffic still needs one. SAP Joule calls it over A2A.

GitHub ↗

follow up

you

What do people say about working with you?

edgar · agent

readcommunity.md, references/5 rows · 3 quotes

render<References />

A few of them said it better than I could: a project manager, a managing partner and someone from my own team. Helping and collaborating with the people around me, so they get to their own goals, is half the fun of this work; the list below is some of it.

<References />rendering

ASUG 2024
SAP BTP and AI sessions for Cogent IBS. Houston (15 Apr), Dallas-Fort Worth (19 Apr) ↗, Denver (20 Sep) and Chicago (4 Nov).
Bootcamps
Co-instructor of Cogent's paid SAP BTP bootcamps. Five sessions in Oct 2023, and the SAP BTP Cloud Bootcamp, 25 Mar to 8 Apr 2025 ↗.
Training
A first course on working with AI day to day has run with a pilot group. I train teams on request; happy to talk about it.
SAP Community
Co-organizer of the SAP Community Mexico group with María Antonieta López and Sandra Solís, since Feb 2026. Profile ↗
Earlier writing
eamarce.wordpress.com ↗, 27 posts from 2015 to 2019 on PI/PO, SAPUI5, PouchDB offline sync and Alexa against SAP.

“rare ability to combine deep technical expertise with practical business execution”

Stephanie WuProject Manager, Cogent IBS, on the automotive program we delivered together. May 2026.

“calm and methodical approach backed by extensive full-stack knowledge”

Phil CoadyManaging Partner and COO, SPOSEA. Managed me directly. Jun 2022.

“one of the best leads I've worked with”

Sainath AllumallaOn my team at Cogent IBS, 2017 to 2019. Mar 2019.

Titles are as they were when each recommendation was written. The full recommendations are on LinkedIn ↗

follow upOpen LinkedIn

you

What should I know about you before we talk?

edgar · agent

readabout.md, principles.mdthe 5 that fit on one screen

readcertifications.json6 SAP certifications · 4 courses on agents

render<Profile />

Seventeen years in enterprise software, most of them on SAP. Below: the facts, and the five principles I actually work by.

<Profile />rendering

NowArchitecture and delivery lead at Cogent IBS for US clients, since 2022

WhereMexico · Remote, US and EU hours; for US and European companies since 2017

LanguagesSpanish native, English C1, German B1

DegreeInformatics Sciences, Instituto Politécnico Nacional, 2007

From an OCR pipeline in 2008 to agents in 2026.

I started in 2008 on an OCR pipeline for a public office, moved to SAP integration and ABAP in 2010, and have worked on SAP's cloud stack since 2015: SAPUI5, then Cloud Platform, then CAP and BTP. I have led distributed teams across India, the US, Mexico and Argentina, up to 16 people between developers, QA and project managers; today the core team is four.

Agents are not my first new thing: SAPUI5 in 2015, Node.js talking to SAP in 2017 when CAP did not exist yet, machine-learning models in 2025. Each time the fun part was the same: understand the pieces underneath, adopt the good practices that already exist, try things on something small, then take them to a client's landscape. That is what makes the skill portable, and that is where I am with agents right now.

Understand first. The problem, the process, the systems, the data, and above all what you want to achieve and what is in the way. I listen before I design.

Objectives, not hours. I build teams where trust runs both ways: everyone owns their part, problems get raised early, and nobody watches the clock, because what matters is that on delivery day we deliver, and deliver well.

Keep a compass. Research what is new, learn the good practices, apply them. And not only on the technical side: a business is after profit and lower costs, so every technical choice has to point there.

Keep things moving. Most delays are not code: they are access, scope, slow decisions and information that does not flow. I work on those every day, and on the relationships that carry them.

Enjoy it. We put our time, our energy and a good part of our lives into this work. It should be worth it.

SAP certifications

2022SAP Certified Development Associate: SAP Extension Suite; SAP HANA Cloud 1.0; SAP Fiori Application Developer
2022SAP Certified Citizen Developer Associate: SAP Build Low-code/No-code Applications and Automations

Earlier: SAP HANA SPS12 Application Associate (2018), SAP Fiori Implementation and Configuration (2016), SAP PI with NetWeaver 7.31 Development Associate (2014), Microsoft Certified Technology Specialist, Web Applications (2014), Oracle SOA Infrastructure Implementation Certified Expert (2011). Credentials ↗

Courses on agents, 2026

AugBecome an Agent-Native Operator (Maven, Aishwarya Naresh Reganti and Kiriti Badam)
SepHarness Engineering: Designing the Operating Layer for Reliable Agents (Maven)
OctBeyond Evals: Designing Improvement Flywheels for AI Products (Maven)
NovBuilding Agentic AI Applications with a Problem-First Approach (Maven, in progress)

Résumé (PDF)LinkedIn ↗GitHub ↗

follow upDownload the résumé

you

Sounds like a fit. What's the next step?

edgar · agent

readways-to-work.mdproject, team or role

render<WaysToWork /> + <Contact />

If you would like to talk, about the field in general or your case in particular, please feel free to send me a message on LinkedIn. I am always happy to connect, and a few lines are enough to start. If a call is easier for you, you are welcome to book 25 minutes or just email me.

<WaysToWork /> + <Contact />rendering

Ways to work with me

Threeways: project, team
or role, full-time or fractional

Since 2022 I lead architecture and delivery at Cogent IBS for US clients. Happy to go over what you have in mind and how we could solve it together.

Mexico · Remote, US and EU hours. Full US business day; European afternoons from 14:00 CET.

Contact

A message on LinkedIn is the easiest way to reach me.

Remote from Mexico; I take calls on US and EU hours.

Message me on LinkedIn ↗Book 25 minutes hello@edgar-martinez.dev

Useful to mention, not required

The system, and what it has to do by when.
What you have tried already.
Whether this is a project, a placement on your team, or a role.

follow upDownload the résumé

SAP BTP architecture and delivery, for teams that need it in production.

Edgar Martinez

Build

Agent assessment

In production

In progress

CT1106 · Zero token architecture for Joule Agents

zero-token-joule-agents

From an OCR pipeline in 2008 to agents in 2026.

SAP certifications

Courses on agents, 2026

A message on LinkedIn is the easiest way to reach me.