MP

Founding Engineer · Cora

Mario Pérez

I design production AI agent systems.

I build the platform agents run on, then put those agents to work in real customer operations.

See the work Email me

About

I spent years shipping software that had to be correct: tax invoicing, health records, cross-border systems, identity. Over the last two years I’ve been building with AI as the default way of working, not as a feature tacked on at the end.

Today I’m a founding engineer at Cora. I design the agent platform and the agents that run on it. The job is to turn messy post-sales work into systems that actually execute: tools, evals, human-in-the-loop, and the data layer underneath.

The bottleneck is no longer writing code. It’s deciding what should exist, then making it reliable.

Based in
California
Working on
Cora: proactive AI agents for customer success
Before
BetterUp · regulated health & tax systems · government

Now

Cora

Founding Engineer · 2025-present

Cora is a platform of proactive AI agents for customer success. Teams connect CRM, email, calendar, calls, and Slack. Cora watches accounts, then runs agents on a schedule or when something happens, to do the work a CSM would otherwise do by hand.

I work both sides of that system: the platform (runtime, skills, signals, evals, permissions, data) and the deployments (taking a new account from zero to steady state). Platform without customers is a demo. Agents without a platform don’t survive production.

  • Agent runtime
  • Evals
  • Skills
  • Data platform
  • Forward deployed
cora.ai

Work

Agent systems in production, then the earlier work that taught me what “correct” feels like.

Customer lifecycle agents

From onboarding follow-ups to QBR prep and save plays: agents that run the recurring work after the sale.

  • Activation rules
  • Email & Slack
  • Human-in-the-loop
  • Evals

Collections & operations

Invoice outreach, ticket context, and human review in the loop: agents that touch money and need a brake pedal.

  • Ticketing
  • HIL
  • Workflow design

Account intelligence

A unified timeline of each account, health and risk signals, and the data layer agents query before they act.

  • Warehouse
  • Signals
  • Integrations

Before agents

FastLane

A secure platform for cross-border passes between Mexico and the U.S. for medical tourists, business travelers, and students.

  • TypeScript
  • .NET
  • JWT
  • Android

BetterUp for Slack

Coaching and development tools inside the flow of work, plus SSO user provisioning with SAML.

  • Ruby
  • PostgreSQL
  • Slack
  • SAML
betterup.com

Electronic invoicing

Cloud invoicing, self-serve kiosks, and a stamping SDK, built to national tax rules, not “best effort.”

  • .NET
  • Java
  • Cryptography
facmail.mx

EHR & health programs

Electronic health records and community-health dashboards: data that cannot be casual.

  • .NET
  • Azure
  • SQL Server

Approach

Decide what should exist

I start from the operation, not the model. If the process is unclear, an agent will only fail faster.

Make it reliable

Tools, evals, permissions, and a human brake. Production agents need the same seriousness as tax and health software.

Close the loop

Deploy with the customer, watch what breaks, feed it back into the platform. That’s the job.

Python · TypeScript · agent evals · data platforms · systems that have to be correct

Contact

If you’re building agent systems that have to hold up in production, write.