Who is Carlos Manuel Cejas?

Carlos Manuel Cejas (also known as Carlos Cejas or cmcejas) is a Cuban-American computer science student at Trinity College Dublin. He is the CEO and cofounder of Lockup, a founding engineer at AVIOM, and a path planning engineer on Formula Trinity. His site is https://cmcejas.dev/.

Where does Carlos Manuel Cejas study?

He studies computer science at Trinity College Dublin in Ireland. He was raised in Philadelphia.

Lockup is a phone-free platform cofounded by Carlos Manuel Cejas. One code puts every phone in the room away. More at https://lockup-app.com.

Carlos Manuel Cejas

Q: What does Carlos Cejas work on?

He builds Lockup (one code puts every phone in the room away), is a founding engineer at AVIOM, writes path-planning for Formula Trinity's autonomous race car, and ships AI agent projects including MCP knowledge bases and open-source agent security work.

Computer science at Trinity College Dublin and Alpine Valley, Cohort Alpha member. Exploring AI agents and MCP knowledge bases, path planning for Formula Trinity's autonomous race car, shipping useful classroom software as CEO and cofounder of Lockup.

Lockup · AVIOM · Formula Trinity

01 Experience

CEO & Cofounder, Lockup April 2026–Present

One code puts every phone in the room away. Lead product, go-to-market, and fundraising while coordinating engineering, design, and customer feedback. Alpine Valley, Cohort Alpha.

site, LinkedIn
Founding Engineer, AVIOM July 2026–Present

site
Path Planning Engineer, Formula Trinity Autonomous Racing Team Oct 2025–Present

Path-planning algorithms in Python for the autonomous race car. ROS/Gazebo simulation and racing-line optimization for higher speeds and shorter laps.
Web Developer, Parkt May 2024–Sep 2024

Led company website development, new features and system optimization. Consulted on branding, operations, and marketing for early-stage growth.

02 Projects

[GoMile] BUILD IRL Vol. 1 · Sep 2026

Won the 1,500 USDT grand prize. Training footage for self-driving is usually just a video. GoMile records the camera together with Sense HAT acceleration, gyro, and orientation, labels what was on the road, and pays the driver in MILE on Solana.

repo, live, linkedin post
[LarpLess] Dublin Demo Day · Oct 2026

A local study-focus tracker using SigLIP 2 image embeddings, OCR and MiniLM text embeddings to classify screen activity. It also detects webcam presence and phone use, and lets you share session stats with friends.

repo, site
[Sol-450] START HACK ST.GALLEN · March 2026

Multi-agent Mars greenhouse simulator with 3D colony view. Bedrock agents and an MCP knowledge base manage crops and resources; DynamoDB for session state.

repo, demo video
[AI Dungeon Master] Claude Builder Club @ TCD Hackathon · Dec 2025

Hackathon-winning AI D&D with persistent campaigns, generated characters, and AI-driven narrative.

repo
[Pollen Cast] Hack Europe · Feb 2026

Allergy health-check app using eye photos, voice, and location. Gemini for vision, Pollen API for allergen data, and librosa for vocal analysis.

repo, live
[Alpine Valley] Alpine Valley, Cohort Alpha · June 2026

Website redesign, brand, and front end rebuilt from scratch for Cohort Alpha, a European early-stage founder program.

live
[Metricare] Collaboration · medical & GDPR compliance · Feb 2026

Built to learn medical and GDPR compliance, and fully compliant with both. Dashboard for patient history, medications, drug interactions, and AI summaries.

repo, live

03 Open Source

[Hermes Agent Security Fix] NousResearch · Security · Open Source · June 2026

Co-authored a merged upstream security fix for NousResearch’s Hermes Agent, a 200k+ star open-source agent framework. The fix closed a serious Telegram auth-ordering vulnerability where removed users could bypass the intended trust boundary long enough to inject prompt-bearing content into an agent’s context before being rejected.

PR #54164
[Hybrid Memory Setup for AI Agents] Open source · June 2026

Open-source memory setup for persistent agents, written for Hermes Agent. Rules that should apply every turn stay in the prompt. Facts sit in a local SQLite archive that checks a hash before a semantic search.

repo

04 Skills

Engineering: Python, TypeScript, React, React Native, FastAPI.

AI: LLM agents, MCP, AWS Bedrock, Gemini.

Infrastructure: AWS, Supabase, Docker.

05 Contact

Email: carlosmcejas@gmail.com

Location: Greater Philadelphia / Dublin

Work authorization: U.S. Citizen with full work authorization in the United States.