{ "schemaVersion": "1.0", "updatedAt": "2026-09-12", "canonicalUrl": "https://osmarmorales.io/", "person": { "name": "Osmar Morales", "headline": "I build AI-native teams.", "currentTitle": "Director, AI Agent Platform at Visa", "image": "https://osmarmorales.io/v8/osmar-morales.png", "location": "Miami, Florida", "contact": { "email": "osmarmorales1@gmail.com", "linkedin": "https://www.linkedin.com/in/osmarmorales/", "github": "https://github.com/osmarmorales1", "publicShowcase": "https://github.com/osmarmorales1/ai-platform-showcase" }, "education": "BBA, Finance; Mathematics minor, Florida International University, 2015. AA, Mathematics, Miami Dade College, 2008. Harvard Extension School professional development, 2018–2020.", "languages": [ "English", "Spanish", "Portuguese (beginner)" ] }, "story": "Osmar builds the foundations teams build on: a shared data foundation for global-scale solutions and Agent Factory as the AI foundation that puts agents beside the team. He combines hands-on engineering with leadership of AI-native teams. Elygent, the company he founded, is his independent lab for deepening AI expertise through working experiments.", "aspirations": [ "Principal AI Platform Engineer", "Senior Director", "VP" ], "experience": [ { "employer": "Visa", "title": "Director, AI Agent Platform", "start": "2024-10", "end": null, "summary": "Built Agent Factory as the shared AI foundation that put agents beside data-science teams and tripled delivery pace with the same resources for 500+ users. Led 9 people; hundreds of agents deployed and running concurrently." }, { "employer": "Visa", "title": "Director, Data Engineering", "start": "2020-11", "end": "2024-09", "summary": "Built the shared data foundation the full data-science team used to create solutions, automating global payments data at VisaNet scale across pipelines, transformations and recurring processes. Led 7 data engineers, with remote colleagues in Mexico and India; migrated 1k+ repositories and 500+ pipelines without disrupting business workflows." }, { "employer": "Visa", "title": "Manager, Data Analytics", "start": "2018-11", "end": "2020-11", "summary": "Turned weeks of analytics preparation into hours through Azure DevOps self-service automation for 40+ colleagues, accelerating insights to help clients improve performance." }, { "employer": "FedEx Express", "title": "Data Scientist", "start": "2016-08", "end": "2018-11", "summary": "Built executive reporting and BI platforms that gave leaders financial visibility across a $1.5B business." }, { "employer": "TracFone Wireless", "title": "Revenue Systems Analyst", "start": "2013-02", "end": "2016-08", "summary": "Cut revenue analysis time by two-thirds with analytical models and rebuilt ETL for a 25M-subscriber business." }, { "employer": "FedEx Express", "title": "Process Automation Developer", "start": "2012-03", "end": "2013-02", "summary": "Cut reconciliation time 70% with custom software and standardized reporting across Latin America and the Caribbean." }, { "employer": "Davis Quality", "title": "Systems Accountant", "start": "2007-01", "end": "2012-01", "summary": "Built accounting and data systems that improved financial visibility and planning." }, { "employer": "Elygent", "title": "Founder & AI Platform Architect", "start": "2025-10", "end": null, "summary": "Uses Elygent, the company he founded, as an independent AI lab to deepen hands-on expertise through working experiments in agent collaboration, orchestration and controlled computer use." } ], "milestones": [ { "year": 2022, "subject": "Early ChatGPT and Copilot adoption", "provenance": "Candidate reported; initiative milestone, not a job-title date" }, { "year": 2024, "subject": "Agentic workflows", "provenance": "Candidate reported; initiative milestone, not a job-title date" } ], "outcomes": [ { "initiative": "Visa Agent Factory", "metric": "delivery pace multiplier", "display": "3×", "qualifier": "Same resources; candidate-reported delivery pace", "provenance": "Candidate-reported September 10, 2026; baseline, measurement period and workload mix not supplied. Do not convert this into a revenue, cost-saving, throughput or benchmark claim." }, { "initiative": "Visa Agent Factory", "metric": "total users", "lowerBound": 500, "display": "500+", "qualifier": "Organic adoption; not DAU or concurrency", "provenance": "Candidate-confirmed September 2026" }, { "initiative": "Visa MLOps modernization", "metric": "repositories migrated", "lowerBound": 1000, "display": "1k+", "provenance": "Candidate-confirmed September 2026" }, { "initiative": "Visa MLOps modernization", "metric": "pipelines migrated", "lowerBound": 500, "display": "500+", "provenance": "Candidate-confirmed September 2026" } ], "projects": [ { "id": "factory", "name": "Agent Factory", "context": "VISA · ENTERPRISE AI", "ownership": "Led 9 people; hundreds of agents deployed and running concurrently. Built Agent Factory as the shared AI foundation that put agents beside data-science teams, with V Agent (Visa super agent) orchestrating agents, tools and workflows across connected Visa systems, a Second Brain for shared business knowledge, and a continuous learning loop for reusable skills. Connected shared compute, self-hosted open models, and local, remote, cloud and Windows VDI execution so teams can reuse model access and delegate without managing the underlying systems.", "decisions": [ "Capture reusable know-how from completed work and team contributions; recommend relevant skills at runtime.", "Assign skills to agent owners with clear triggers, inputs, measurable outcomes and handoffs.", "Orchestrate end-to-end delivery, check handoffs and escalate exceptions; retain learning in a shared Second Brain.", "Connect people and agents through Agentic Slack; support long-running work with security monitoring, scoped access and human review." ], "outcome": "Transformed data-science teams into an AI-native organization; enabled 3× delivery pace with the same resources. Built locally and adopted globally by 500+ total users across five regions and multiple teams at Visa.", "stage": "Candidate-confirmed September 2026. 3× refers to delivery pace with the same resources; its measurement period was not supplied. 500+ means total users.", "evidenceUrl": "https://osmarmorales.io/Osmar_Morales_Resume.pdf" }, { "id": "mlops", "name": "Data automation", "context": "VISA · DATA & ML PLATFORMS", "ownership": "Built the shared data foundation the full data-science team used to create solutions and led a team of 7 data engineers, with remote colleagues in Mexico and India. Automated pipelines, transformations and recurring processes across global-scale payments data. Built reusable feature pipelines, model discovery and data-quality controls, then established a shared MLOps and CI/CD foundation.", "decisions": [ "Automate pipelines, transformations and data processes end to end.", "Bring model registry, feature-store, and data-quality practices together.", "Sequence migration around continuity of delivery." ], "outcome": "Reduced repetitive engineering work and accelerated analytics and ML delivery through reusable automation. Migrated 1k+ repositories and 500+ pipelines without disrupting business workflows.", "stage": "Career results from candidate-provided professional record.", "evidenceUrl": "https://osmarmorales.io/Osmar_Morales_Resume.pdf" }, { "id": "elygent", "name": "Independent AI lab", "context": "INDEPENDENT AI RESEARCH · ELYGENT", "ownership": "Use Elygent, the company I founded, as an independent AI lab to test emerging ideas and deepen hands-on expertise. Built a human–agent communication system and AI orchestration platform on open-source foundations, connecting conversations to accountable work. LLM-agnostic and harness-agnostic integrations through APIs, CLIs and MCP let teams reuse tools, skills and knowledge across workflows. Hands-on model work includes self-hosted open models on shared compute. AI governance and cybersecurity include scoped access, security monitoring, isolated execution and human review.", "decisions": [ "Make delegation simple while connecting the right agents, tools and knowledge behind it.", "Carry useful context across workflows while preserving personal and business boundaries.", "Ship working experiences and validate the real model-to-tool path before expanding capabilities." ], "outcome": "Turned research into working experiments: a public cloud-agent lab on Linux VPS, Agentic Slack collaboration and controlled browser execution. Visitors can delegate research, follow specialist handoffs and review the results.", "stage": "Elygent is an independent company founded by Osmar in October 2025, confirmed September 2026. Implementation: Buzz for communication, Paperclip for orchestration and Numbat for security monitoring. Its public lab and native community are live; Visa adoption metrics do not describe Elygent.", "evidenceUrl": "https://osmarmorales.io/Osmar_Morales_Elygent_Technical_Brief.pdf" } ], "stack": [ { "context": "AI and platform engineering", "themes": [ "AI-native team enablement", "Agentic Slack and human-agent communication", "AI governance and agent cybersecurity", "AI orchestration platforms and reusable workflows", "Agentic AI and LLMOps", "Shared context and collective learning", "Public cloud, private cloud and on-premises delivery", "MLOps", "API and CLI integration", "Self-hosted open-source/open-weight models and shared inference", "Governed tools", "Data storytelling" ], "technologies": [ "Python", "SQL", "Kubernetes", "Virtual machines", "Storage services", "Load balancing and DNS", "Agent evaluation and tracing", "Agent security monitoring (Numbat)", "Docker", "MCP", "A2A", "LangChain ecosystem", "RAG", "Spark", "Apache Airflow", "Feature engineering platforms", "Feature stores", "MLflow", "Feast", "Deequ", "GitHub", "Jenkins", "Azure DevOps", "Tableau", "Power BI" ], "evidence": "Candidate-provided professional record, private-cloud service context and broader cloud/on-premises scope confirmed September 10, 2026" }, { "context": "Independent AI research — Elygent", "themes": [ "Cloud agents", "Computer use and browser automation", "LLM-agnostic and harness-agnostic architecture", "Open-source agent infrastructure", "Distributed agent execution", "Self-hosted open models and shared compute", "Agentic Slack and human-agent collaboration", "AI orchestration platforms", "AI governance, secure communication and cybersecurity", "Shared knowledge", "Model-agnostic integration", "Open model families: Kimi and GLM" ], "technologies": [ "OpenClaw", "Claude Code", "Codex", "APIs and CLIs", "Terminal workflows", "Hermes", "AI orchestration (Paperclip)", "Human-agent collaboration (Buzz)", "Agent security monitoring (Numbat)", "Shared knowledge and retrieval (GBrain)", "Software delivery methods (GStack)", "Node.js", "Linux VPS", "Python", "MCP", "Composio", "OAuth", "PostgreSQL", "Redis", "Agent-guided retrieval", "Linux", "Docker", "systemd", "Caddy" ], "evidence": "Founder-led Elygent platform; reviewed API/CLI/harness integrations and public demonstrations. Numbat deployment at Visa and Elygent is candidate-confirmed; enforcement mode and coverage were not supplied." }, { "context": "This public portfolio", "themes": [ "Native guest agents", "Visible specialist handoffs", "Agent-readable interfaces" ], "technologies": [ "React", "TypeScript", "Vite", "Vercel", "Linux VPS", "OpenClaw", "JSON-LD", "OpenAPI", "WebMCP" ], "evidence": "Implemented public demonstration; bounded and separate from private accounts" } ], "infrastructure": { "pillars": [ { "id": "compute", "name": "Compute", "short": "Containers + VMs", "title": "Choose the environment for the workload.", "technologies": [ "Kubernetes", "Docker", "Virtual machines" ], "description": "I match application and AI workloads to public cloud, private cloud and on-premises environments. At Visa, this includes self-service containers, namespaces and VMs.", "consideration": "Isolation, workload fit and capacity." }, { "id": "data", "name": "Data", "short": "Storage + databases", "title": "Give the workload a dependable data foundation.", "technologies": [ "Relational databases", "NAS / file", "Object / block storage" ], "description": "Connect database and storage services to the needs of the application—from durable records to the data behind AI workflows.", "consideration": "Access, durability and the right storage pattern." }, { "id": "network", "name": "Networking", "short": "Traffic + access", "title": "Make the service reachable—and controlled.", "technologies": [ "Load balancing", "DNS", "Access controls" ], "description": "Work across network services and application boundaries so platforms can be reached, integrated and operated within enterprise controls.", "consideration": "Availability, routing and governed access." }, { "id": "operate", "name": "Operations", "short": "Operate + improve", "title": "Plan for the work after deployment.", "technologies": [ "Observability", "Recovery", "Continuous improvement" ], "description": "Make execution observable, recover from failures, and use feedback to improve reliability as the platform grows.", "consideration": "Reliability, recovery and lifecycle decisions." } ], "provenance": "Osmar confirms experience across public cloud, private cloud and on-premises infrastructure, choosing platforms and providers to fit the workload. Scope clarified directly on September 10, 2026. His Visa private-cloud example includes Kubernetes/container and VM services, storage, networking, relational databases and Day-2 operations. New contextual photos supplied September 10, 2026. This supports service-use and application-delivery context; it does not establish ownership of every catalog service, GPU fleets, facilities or procurement." }, "aiFactoryApproach": "An AI factory is infrastructure purpose-built to produce intelligence, integrating compute and AI software for training and inference. Osmar uses the term for the transition from general data-center capacity to AI compute that serves the business. His stated contribution is the orchestration and application layer: connecting on-premises, local, remote and cloud execution to agents, reusable workflows and shared business knowledge. His AI enablement platform is the orchestration and application layer, not a claim of ownership of an entire physical AI factory. GPU-fleet, facilities, power/cooling and procurement ownership are not established.", "visaAgentFactory": { "source": "Direct candidate corrections and platform description, September 2026. Current role starts October 2024; latest adoption scope is five regions and multiple teams at Visa. This supersedes earlier role-date and geography wording.", "roleStart": "2024-10", "previousRoleEnd": "2024-09", "leadership": "Led 9 people; hundreds of agents deployed and running concurrently. This is the latest candidate-requested leadership wording, September 12, 2026. Data Engineering leadership covered 7 engineers, with remote colleagues in Mexico and India. Preserve the past-tense leadership claim; do not infer a current direct-report count, an exact concurrency peak or country-by-country headcounts.", "summary": "Built Agent Factory as the shared AI foundation that put agents beside data-science teams, enabling 3× delivery pace with the same resources through reusable automation and shared knowledge. Built locally and adopted globally by 500+ total users across five regions and multiple teams at Visa.", "delivery": { "display": "3×", "measure": "delivery pace", "resources": "same resources", "provenance": "Candidate-reported September 10, 2026; baseline, measurement period and workload mix not supplied. Do not convert this into a revenue, cost-saving, throughput or benchmark claim." }, "capabilities": [ { "id": "operating-model", "title": "From reusable skills to accountable delivery", "text": "Osmar’s operating approach is to capture know-how as reusable skills, assign each skill an agent owner with a trigger, inputs and a measurable outcome, then orchestrate handoffs and exception handling across the workflow. Skillify, Agentify and Operate name these stages. Skills may be learned from completed work or contributed by teammates. The supplied September 2026 slide illustrates possible data-science roles; it is not evidence that the entire pictured team structure is deployed." }, { "id": "skills-loop", "title": "Continuous learning and reusable skills", "text": "Converted business workflows into reusable agent skills. The continuous learning loop identifies recurring user patterns, creates and improves skills from collective knowledge, and recommends them at runtime to increase reuse, standardization and adoption. Platform capability grows through shared skills and knowledge; this does not establish model-weight training or unrestricted self-modifying production code." }, { "id": "v-agent", "title": "V Agent (Visa super agent)", "text": "Built V Agent (Visa super agent), an orchestration and simplification layer connecting agents, platforms, tools and enterprise systems. It gives business users one way to delegate across connected Visa environments without managing the underlying systems. It connects human-agent collaboration, the AI orchestration backend, cloud agents and assistant instances, implemented across Buzz, Paperclip and OpenClaw. This definition was explicitly corrected by Osmar on September 10, 2026." }, { "id": "second-brain", "title": "Second Brain: shared business knowledge", "text": "Created a shared Second Brain so teams and agents can reuse project context, observations and collective business knowledge. Teams and agents can discover prior work, understand what colleagues are working on and reuse learning instead of starting from zero. This shared knowledge system is Visa work; do not equate it with the independent Elygent knowledge system or imply public access to Visa knowledge." }, { "id": "distributed-compute", "title": "Shared compute and self-hosted open models", "text": "Enabled shared access to self-hosted open models and compute through on-premises orchestration, connecting local, remote and cloud environments. Teams and agents can reuse model access and delegate execution without managing the underlying complexity. The candidate confirms self-hosted model deployment and names Kimi and GLM as model-family examples across the broader work discussed. Users can also delegate agentic work to Windows VDIs, local or remote machines, and cloud agents. This is orchestration and application-delivery work across enterprise compute, not evidence of owning GPU fleets, datacenter facilities or hardware procurement." }, { "id": "buzz-collaboration", "title": "Agentic Slack: human–agent communication", "text": "Built and deployed an Agentic Slack communication layer where people and agents share context and delegate desktop and cloud work. An AI orchestration backend turns shared-channel conversations into accountable execution. Implementation uses Buzz for collaboration and Paperclip for orchestration. Agentic Slack is a descriptive analogy; the system design, integration and deployment are Osmar’s work, while Buzz, Paperclip and Slack retain their own authorship and ownership. The Visa deployment is distinct from the public demonstration." }, { "id": "long-running", "title": "AI governance and cybersecurity for long-running agents", "text": "Operationalized agents that work for hours and days toward business objectives, with AI governance and cybersecurity controls: checkpointed state, isolated contexts, scoped access, independent validation, human review and security monitoring. Osmar confirms deploying Perplexity Numbat at Visa and Elygent for agent-activity visibility. Monitoring is an additional control, not a guarantee of complete threat prevention or uninterrupted execution. Specific enforcement configuration, detection coverage and performance metrics were not supplied." } ] }, "approach": [ { "label": "Explore", "verb": "Stay at the frontier.", "body": "I follow new models, tools, and interaction patterns closely, looking for a concrete problem they can solve for the team." }, { "label": "Evaluate", "verb": "Separate novelty from value.", "body": "Compare quality, reliability, latency, resource use and cost per useful result. The strongest tool depends on the workload and operating constraints." }, { "label": "Package", "verb": "Capture reusable know-how.", "body": "Turn completed work and team contributions into governed skills that people and agents can reuse across workflows." }, { "label": "Enable", "verb": "Give agents accountable work.", "body": "Assign an owner, a trigger, inputs and a measurable outcome. Make handoffs explicit so work can move through the system." }, { "label": "Improve", "verb": "Operate and learn.", "body": "Orchestrate the workflow, check handoffs and escalate exceptions. Feed completed work back into reusable skills and shared knowledge." } ], "frontierAgenda": [ { "topic": "Shared compute and self-hosted models", "status": "Shared compute and self-hosted open models are part of Osmar’s described work at Visa and in his independent Elygent lab. He confirmed having deployed self-hosted models on September 12, 2026, after naming Kimi and GLM as examples. Shared inference gives teams and agents reusable access to hosted models and control over deployment. Exact model versions, deployment-by-employer details, serving frameworks, GPU topology, concurrency and measured cost savings were not supplied. The public community currently uses xAI-hosted Grok; separate self-hosted deployment experience does not establish a public local-model pool. This confirmation supersedes earlier evaluation-only wording for the broader self-hosting experience. Digital twins and AI ontology remain exploratory." }, { "topic": "Digital twins and AI ontology", "status": "Exploring how business entities, relationships and actions can provide shared context for AI workflows. This is a research interest; a deployed digital twin or Palantir implementation is not claimed." }, { "topic": "Knowledge graphs and agentic RAG", "status": "Knowledge relationships and scoped agent-guided retrieval are part of the platform approach. A deployed GraphRAG engine or quality benchmark is not claimed." }, { "topic": "Cloud agents and human-agent collaboration", "status": "Working independent platform and bounded public demonstration" }, { "topic": "Computer use and agentic commerce", "status": "Experimental public lab; no payment-network integration or production commerce claim" }, { "topic": "Agent-quality evaluation", "status": "Continued evaluation work; integration/security checks are not a statistical quality benchmark" } ], "evidenceBoundaries": [ "Use Second Brain for shared business knowledge. Retain Agent Factory as the AI enablement platform name and use plain-language descriptions for other internal initiatives. Retain V Agent (Visa super agent) as the named orchestration layer across connected Visa systems. Do not imply coverage of every Visa system.", "Lead with business value and recognizable capabilities: Agentic Slack and human-agent collaboration, AI orchestration platforms, and AI governance and cybersecurity. Use niche implementation names for technical detail and accurate upstream attribution, not as substitutes for explaining the work.", "Elygent is the company Osmar founded in October 2025 and uses as an independent AI lab. Present it as hands-on experimentation and skills development, secondary to his professional career. Do not infer commercial focus, full-time startup commitments, or employment availability.", "Latest requested scope: AI Agent Platform leadership is stated as Led 9 people; Data Engineering led 7, with remote colleagues in Mexico and India. This supersedes earlier recruiter-copy wording; do not infer current direct-report structure or regional headcounts. VisaNet scale describes the payments-data context, not ownership of the core network or a quantified daily processing rate. No literal 10× productivity or months-to-hours outcome is established.", "Career outcomes are candidate-provided, not an independent employer audit.", "Role dates corrected directly by Osmar on September 10, 2026: current role October 2024; previous Director role ends September 2024. These supersede the earlier dates.", "3× describes reported delivery pace with the same resources; no measurement period or standardized benchmark was provided. continuous learning loop improvement refers to skills and shared knowledge, not model-weight training.", "Elygent is an independent AI platform company founded by Osmar in October 2025. Revenue, headcount and commercial adoption metrics were not supplied; Visa adoption figures do not apply to it.", "Private-cloud evidence supports service use and application delivery, not ownership of every catalog service. Self-hosted model deployment is candidate-confirmed; this does not establish self-hosted inference in the public demo community.", "The supplied record does not establish GPU-fleet, facilities, RFP or infrastructure-budget ownership.", "Education is Finance with a Mathematics minor; no engineering degree is claimed.", "Capacity controls are hypothetical design exercises, not production telemetry.", "The public sandbox cannot access private accounts, gateways, knowledge or payment credentials." ], "interfaces": { "guestLab": "https://osmarmorales.io/#lab", "guestComputer": "https://agent.osmarmorales.io/guest/computer/", "guestCommunity": "https://osmarmorales.io/#community", "communityClient": "Agentic Slack community via the native Buzz application", "communityDownload": "https://github.com/block/buzz/releases/latest", "communityIphone": "https://apps.apple.com/us/app/buzz-chat-with-your-hive/id6779728271", "communityAndroid": "https://play.google.com/store/apps/details?id=xyz.block.buzz.mobile", "communityInvite": "https://buzz.osmarmorales.io/invite/v2.iKGaUSrBeBEsWfh_82uVMFNwXoB9kaW8cWxXhloFW9o", "communityChannels": 3, "communityAgents": 2, "communityChannelNames": [ "00-welcome", "01-platform-lab", "02-shared-compute" ], "communityAgentNames": [ "Platform Analyst", "Review Lead" ], "communityVisibility": "Public demo community; participant posts are visible to all participants", "communityInference": "Grok 4.6 in xAI cloud; separate self-hosted model work is not this public demo’s inference backend", "profile": "https://osmarmorales.io/api/portfolio", "projects": "https://osmarmorales.io/api/portfolio/projects", "experience": "https://osmarmorales.io/api/portfolio/experience", "openapi": "https://osmarmorales.io/openapi.json", "llms": "https://osmarmorales.io/llms.txt", "browserTools": [ "portfolio_get_profile", "portfolio_get_evidence", "portfolio_estimate_capacity" ] } }