Origin Intelligence
I'm Luis Sanchez Guzman. Origin Intelligence is the company through which I take on engineering work: architecture, integrations, automation, testing and operations. Most of it involves Salesforce, AI agents or the tools developers use every day.
It suits teams that already run on Salesforce or already use AI agents, and need one person who can design a change, build it and show that it works.
What I do, and where I have done it
Salesforce commerce and customer data
Features, integrations, fixes and Agentforce agents on Salesforce Commerce Cloud and Data Cloud. I start from what your org actually contains, test against a sandbox and leave production untouched until you approve a change.
- Data Cloud ingestion, identity resolution, calculated insights and activation
- Agentforce agents that answer from your CRM and Data Cloud data, with scoped topics and actions
- SFRA and PWA Kit storefronts, and internal apps on Managed Runtime
- Integrations with payment gateways and back-office systems
- Security reviews against the Security Benchmark for Salesforce
Inventory health dashboard
A dashboard on Salesforce Managed Runtime for Dabra's retail brands Dexter, Moov and Stockcenter. It shows stock health across the brands, flags anomalies before items run out and reports commercial figures from Data Cloud, behind the retailer's own sign-in. It replaced reconciling three spreadsheets by hand.
Data Cloud knowledge base
A knowledge base, maintained with an LLM, for Dabra's production Data Cloud implementation. It keeps Salesforce's documentation next to the decisions recorded in that org, such as which join keys are safe and what an identity resolution rerun costs, so answers come from the org and not from generic advice.
Analytics agent on Agentforce
An Agentforce employee agent that answers the operations and growth teams' questions about daily sales, trends, cancellations and why a metric changed, for Dexter, Moov and Stockcenter. It answers only through Apex actions that query Data Cloud calculated insights, states the date and currency of every figure, and declines questions outside its topics. A 10-question test set, in English and Spanish, checks the topic, action and figures it returns.
Salesforce Security Benchmark for Pi
A package for the Pi coding agent. It detects security-relevant Salesforce files as the agent works, surfaces the applicable controls among the benchmark's 25, and adds audit, fix and review workflows.
Agents and workflow automation
Agents and workflows that take repetitive work off a team without taking its decisions. They run where the team already works, such as Slack, Jira, n8n or Salesforce, and ask a person before any write that matters. Each one ships with an evaluation set, so you know how often it is right before you rely on it.
- Classification and routing of incoming requests
- MCP servers that give agents controlled, logged access to your data
- Evaluation sets and execution logs for agents in production
Request triage in Slack
When someone mentions a product squad in Slack, a workflow classifies the message as a bug, question, feature request, incident or other and posts it to a triage channel. When a ticket is needed, it drafts one and creates it in Jira only after a person confirms in the thread. Classification is measured against a 30-message evaluation set. In production at FCM since March 2026.
Daily SEO agent for this site
Every morning it reads this site's analytics, search queries, HTML and Lighthouse results, and proposes changes as a pull request. On its first run it found a Spanish phrase in the structured data that I had missed. It also once proposed preloading an image that does not exist, which is why a person reviews every change.
Coding agents for engineering teams
Setting up coding agents so that their work can be reviewed and trusted: repository instructions, reusable skills and checks that fail when something regresses. I work this way every day, including on this site.
- Instructions and skills for Claude Code, Amp and Pi
- Verification in CI: screenshot comparison, Lighthouse budgets and accessibility checks
- Extensions that connect agents to developer tools, such as Figma
Company memory for agents
A memory layer that gives agents persistent context about a company: decisions, facts and past work, retrieved at the start of each session instead of re-explained. I use it every day with my coding agents. It is an MCP server with keyword and semantic search over SQLite, OAuth 2.1 sign-in and more than 590 tests. The same pattern fits a team whose agents keep asking for context it has already given.
Figma reader for Pi
A Pi extension that lets an agent read Figma files without filling its context. It reduces each API response to a compact shape, 84.9% smaller on a measured frame, and caches by the file's version.
How an engagement works
Diagnostic
A short, paid diagnostic with an agreed scope. I learn how your team works and what your systems actually contain. It ends with a written plan: what to build first, the risks, and which parts your team can build without me.
Build
I quote the build from the plan, then build the first piece in your repositories and sandboxes, with tests and a written way to verify it. I change production only after you approve a specific change.
Handover
The code, its documentation and its checks already live in your repositories, in a state your team can run and change. If you prefer, I keep operating it.
Background
I started writing software in 2007, then moved into managing digital projects and consulting teams in Latin America and Spain. From 2019 to 2022 I ran multi-country e-commerce programs for consumer-goods companies at OSF Digital. From 2022 to 2024 I led a practice of Salesforce consultants and architects at VISEO. I founded Origin Intelligence in August 2024.
- Salesforce Certified Data Cloud Consultant
- Salesforce Certified Business Analyst
- Salesforce Certified AI Associate
- Salesforce Certified Associate
- Project Management Professional (PMP), Project Management Institute
Questions
Can an agent work inside our existing Salesforce org?
Yes. An agent connects through Salesforce's standard APIs, an MCP server or platform events, and runs as a user whose permissions you control. It can read and update records, start flows and classify incoming cases. I build and test against a sandbox first.
What does an agent do without a person?
Only work that is easy to check or undo, such as reading, classifying and drafting. Writes that matter, like a new ticket, a changed record or a published page, wait for a person to confirm. The SEO agent on this site opens pull requests; it never publishes.
How do you know an agent is good enough?
Before it goes live, I run it against real examples with known answers and report how many it gets right. In production it logs each decision, so wrong answers can be found and added to the examples.
What is an MCP server, and when is one worth building?
MCP (Model Context Protocol) is an open standard that lets an AI model call tools and read data through a defined interface. An MCP server exposes a small set of operations, such as a search or a record lookup, with its own sign-in and logs. It is worth building when several agents or people need the same access and you want to control and audit it in one place.
Which industries do you work with?
Mostly retail and consumer goods, where I spent most of my years in Salesforce commerce. The same work fits any team with structured data, repeated decisions and more requests than people.
Where are you based, and in which languages do you work?
Origin Intelligence LLC is a US company, registered in Delaware, and works with clients anywhere, remotely. I am based in Madrid and work in English or Spanish.
Contact
Write to me with what you want built or fixed, and what you have already tried.