- Multi-agent orchestration across Claude Code, Codex and Cursor under one Spec-Driven workflow.
- Engram replaces full-context replays with retrieval-augmented agent state: ~10,200 startup tokens cut per agent invocation.
- ~40% lower total token consumption with improved task continuity across sessions.
- Adopted across the software factory as the standard development workflow.
[ SR/2026 ] - Portfolio Guadalajara, MX · Field notes of a forward-deployed engineer
Sebastian Rojas
Engineering Manager · GenAI & Multi-Agent Systems Architect. I design and ship production agent systems for government and enterprise.
[ 001 ] - Summary
Agents in production, not demos.
Engineering manager and GenAI architect delivering production systems for government and enterprise clients, including multi-agent GenAI deployments.
I design and ship developer harnesses, agent memory layers and Spec-Driven Development workflows that measurably cut token consumption and accelerate delivery.
Beyond agents: I have shipped B2B enterprise integrations (EDI 214/810, SAP CPI, AS2/SFTP), serverless image-classification APIs, multi-carrier insurance quoting platforms and AI-assisted incident detection systems — across AWS, Azure and GCP.
[ 002 ] - Experience
Deployments, not case studies.
Click a card to uncover the details.
Head of Software Factory
Jul 2024 - Present- OpenSpec-style workflows and spec-first handoffs standardized across teams.
- Reduced ambiguity at delivery boundaries; development velocity up 20%.
- Government of Guadalajara: conversational agent with multi-MCP server integration, RAG over document repositories and live API connectors for real-time retrieval.
- Government of Michoacán: central orchestrator with two sub-agents, one querying an Azure Data Lake via MCP (ingesting from AWS MySQL), the other retrieving vectorized documents from a Redis vector store.
- EDI 214/810 and XML over AS2/SFTP, orchestrated with Azure Functions, Logic Apps and SAP CPI.
- Diagnosed mTLS chain-of-trust issues across InforNexus and Azure Key Vault.
- Defined the Software Delivery Engineer profile: Domain Immersion, Trusted Advisor, Product Engineering Mindset, BizDevOps translation.
- Led ISO 20000 and ISO 27001 renewals; contributed to AWS DevOps Competency Partner status.
Software Factory Team Lead
Nov 2023 - Jul 2024- XML web-service integrations with multiple insurance carriers.
- Role-based access and post-sale tracking.
- Cross-platform app with Google Analytics instrumentation, increasing customer engagement.
- Serverless API for automated product image classification, improving marketplace security and seller onboarding.
Software Engineer
Jul 2023 - Nov 2023- Cloud platform on Flask, React + Vite and Docker, integrating AWS Rekognition, Textract and Detectron, with Tableau analytics.
- Streamlined static vehicle content management for JAC Motors Mexico, reducing time-to-publish.
Production Engineering Fellow
Jun 2022 - Aug 2022- Python, Flask, Jinja, MySQL and Nginx.
- Automated CI/CD pipelines; monitoring with Prometheus and Grafana.
[ 003 ] - Side ventures
Built on the side.
An autonomously-operated consulting studio.
A Jalisco-based technology consulting studio focused on tailored software, consultative engagement and short delivery cycles.
- Operations run autonomously on a custom OpenClaw multi-agent harness with specialized chief-of-staff agents.
- Notion-based CRM handles lead intake, qualification and follow-up with minimal human-in-the-loop.
- Infrastructure on Cloudflare and AWS; design, project management and engineering under one roof.
[ 004 ] - Skills
The stack.
GenAI / Agents
Agent systems, end to end.
- f1
- f2
- f3
- f4
- f5
- f6
- f7
- f8
- f9
[ 005 ] - Credentials
Certified & schooled.
Certifications
AWS DevOps Engineer Professional
i_awsAWS Developer Associate
i_awsAWS Cloud Practitioner
i_awsProfessional Cloud Developer
i_gcpAzure Developer Associate
i_azITIL V4 Foundation
i_itilEducation
Instituto Tecnológico y de Estudios Superiores de Monterrey
B.Sc. Computational Technologies Engineering
4.0 / 4.0 GPA
[ 006 ] - About
Beyond the terminal.
Engineering is what I do. These are some of the things that shape how I think.
ab_g1_t
ab_g1_p
ab_g2_t
ab_g2_p
ab_g3_t
ab_g3_p
ab_g4_t
ab_g4_p
[ 007 ] - Contact
Let's build agents.
Open to conversations about multi-agent systems, GenAI architecture and engineering leadership. The fastest way to reach me is email.