About the Founders
The people behind Joltrin
Joltrin has two founders. Gerardo Recinto wrote the first commit in 2015 and is the main author of the code. Gerard Louis Recinto is a DevOps and platform engineer who built the MCP and A2A servers, the Kubernetes admission gate and the delivery pipeline.
Gerard Louis Recinto
Founder of Joltrin
I have spent my career building the automation that engineering teams depend on. Joltrin brings the same habits to AI agents: check the step before it runs, and say what to do first.
Background
I am a Senior DevOps and platform engineer based in San Diego, California. I build the pipelines, cloud infrastructure and Kubernetes platforms that other engineers ship on, and I spend a lot of time troubleshooting them in production.
At Qualcomm that work has included supporting more than 475 Jenkins pipelines, cutting one CI build from 6 hours 59 minutes to about 15 minutes by moving binary artifacts from Git LFS to Artifactory and writing a Python 3 multiprocessing script as part of that change, and implementing an S3 lifecycle change that saved about $26 million a year, a figure reported by our AWS technical account manager.
I studied Mathematics and Computer Science at UC San Diego and hold the AWS Certified Solutions Architect Associate certification. My day-to-day tools are Python, Terraform, Kubernetes, AWS and CI/CD systems.
Those results come from my professional work. They are not Joltrin customer results, and no employer or vendor endorses Joltrin.
What I built in Joltrin
I have written about 445 commits in this repository, and not only the pipeline around the code. These are the parts that agents and clusters actually talk to:
- MCP server.
sop-mcp-serverexposes the verification engine to agents over the Model Context Protocol. It installs with one line, ships as small prebuilt binaries on every release, and registers itself by full path so a missing PATH entry cannot break it. Blocked and malformed requests get structured, actionable errors. - A2A agent and bridge. An Agent2Agent endpoint over the same engine, its agent card, and an MCP-to-A2A bridge with tests and setup docs.
- Kubernetes admission gate.
sop-k8s-gateis a validating admission webhook written with the standard library only, so the core module gained no dependencies. The API server refuses a delete until the steps before it are recorded for that trace. It reloads its certificate on rotation and ships with cert-manager manifests, Prometheus counters, an audit log, a deny-egress network policy, a signed and scanned image build, a Kyverno image check in the AKS demo, and a script that reproduces the checks on a real kind cluster. - Clients and examples. A LangGraph agent over MCP that recovers from a block, with its own CI run, and a Rust client for the gate command.
- Delivery. The race, coverage and staticcheck gates, release pipeline, site deploy and rollback that ship all of the above.
The commit history is public in the repository, so you can check each of these. Gerardo wrote most of the core storage code these sit on.
Languages and skills I use here
- Go
- The MCP and A2A servers, the bridge and the Kubernetes admission webhook, plus the race, coverage and staticcheck gates that test them.
- Rust
- A Rust client for the gate command, with tests and an example. I also fixed the Rust build on Apple Silicon, where a Go toolchain running under Rosetta built the wrong architecture.
- Python
- A LangGraph agent that talks to the MCP server and recovers from a block, with its own CI run, and the Python library loader and its test script.
- AI agents
- Model Context Protocol, Agent2Agent, and guardrails that block a risky agent step until the steps before it are recorded.
- Kubernetes and cloud
- Admission control with cert-manager, Prometheus metrics, network policy, Kyverno image checks on AKS, and a kind cluster test.
- Delivery and testing
- GitHub Actions, signed and scanned container images, Cloudflare deploys with rollback, and TypeScript Playwright tests in three browsers.
At a glance
- Based in
- San Diego, California
- Role
- Senior DevOps and platform engineer
- Education
- B.S. Mathematics and Computer Science, UC San Diego
- Certification
- AWS Certified Solutions Architect Associate
How that shapes Joltrin
Stop the risky step, then say what comes first.
A release pipeline does not just fail a bad deploy. It tells you which check to run. Joltrin works the same way: it blocks a risky agent step, like dropping a database, until the steps before it are recorded, and tells the agent what to do first.
Let people try it before they trust it.
You can run the barrier in your browser, and the core is open source under the MIT license, so you can read exactly what it does.
Keep the workflow readable.
The automation I have seen last is the kind a new engineer can follow in an afternoon. I try to keep Joltrin's rules and checks that plain.
Gerardo Recinto
Founder of Joltrin
Background
Gerardo is a software engineer based in Union City, California. He made the first commit of this project in April 2015, when the repository was named SOP, for Scalable Objects Persistence, and he has written most of the code since. As of November 2026 that is more than 1,900 of the repository's roughly 2,400 commits.
The project began as a Go library for storing objects durably, built around a transactional B-tree. It is now durable memory and a verification barrier for AI agents in one embedded Go library, and the B-tree is still at its core. You can try the B-tree yourself in the B-tree game.
Gerardo also publishes TensorTree, a vector database in Go built on the same B-tree.
- Based in
- Union City, California
- Role
- Founder and main author of the code
- On GitHub
- github.com/grecinto
Everything above comes from the public commit history and Gerardo's public GitHub profile.
See how it works
Try to make an agent drop a database and watch what the barrier does.