Case Study | Agent Platform

Enterprise MCP platform for governed AI workflows.

A production orchestration layer connecting Claude to NetSuite, Monday.com, HubSpot, Wrike, Google Ads, Adverity, Apify, Snowflake, LinkedIn, Jasper, custom Claude skills, and human approval workflows.

Discuss Similar Work Download Resume

Agent platform artifact

Enterprise systems become tools Finance, CRM, project management, advertising, data warehouse, professional network, and content platforms behind governed MCP contracts.

OAuth and approval boundaries Auth0, Google Ads, LinkedIn, Jasper, and Snowflake paths designed around authentication, access control, and human review.

Reusable Claude skills 23 skills encode repeatable strategy, research, approval, and content operations.

458 tools 23 skills 4 Jasper clients

Hiring signal

This case study shows agent engineering beyond chat UX: tool design, workflow boundaries, reusable skills, cross-platform orchestration, and controlled autonomy.

Best role fit: Principal AI Platform Engineer, Staff Agent Systems Engineer, Applied AI Lead.
Operating mode: package domain IP into reusable tools, skills, and governed workflows.
Risk profile: brand quality, content review, per-tenant context, tool use, and human-in-the-loop controls.

458 MCP-powered production tools across agent and content workflows

23 packaged Claude skills encoding B2B marketing methodologies

34 page Monday.com MCP assessment covering feasibility, security, licensing, and rollout

4 enterprise B2B clients supported through the custom Jasper content workflow

Artifact Packet

What a reviewer should inspect first.

These are the kinds of artifacts that separate a governed agent platform from a demo wired to a few APIs.

Registry sample

Tool contract inventory

Names, scopes, owners, auth modes, approval needs, rate limits, and failure behavior for agent-callable enterprise tools.

Governance checklist

Scope-diff review

A change review path for new tool powers, permission creep, account boundaries, and actions that should require human approval.

Approval object

Human-in-loop schema

Structured review requests containing action intent, source data, confidence, blast radius, reviewer identity, and audit trail.

Incident note

OAuth recovery log

LinkedIn/Auth0 token diagnosis, audience parameter root cause, JWT recovery, and downstream account-access validation.

Architecture

A governed agent platform, not a pile of prompts.

The platform packaged enterprise SaaS integrations, reusable business methods, and internal knowledge surfaces as governed tools and skills, then kept quality-critical decisions behind authentication, access, and human approval gates.

System flow

brief to approved assets

Inputs

Strategic briefAudience, goals, offer, constraints, channel plan, and source material.

Enterprise systemsNetSuite, Monday.com, HubSpot, Wrike, Google Ads, Snowflake, LinkedIn, and Jasper.

Agent layer

Claude workflowsResearch, planning, structured JSON briefs, content pack generation, and operator assistance.

Jasper bridgeCustom content-pack-pipeline server connecting Claude to Jasper Knowledge Base and brand voice selection.

Tool layer

MCP registry458 tools connecting data stores, workflows, approvals, platform services, and enterprise SaaS.

Governed authAuth0 OAuth, LinkedIn JWT recovery, Google Ads account access, and Snowflake role controls.

Outputs

Content and reportingBriefs, content packs, finance reconciliation, audience intelligence, and platform assessments.

Review gatesHuman approval at quality-critical junctions before content, spend, or reporting changes ship.

What I owned

Architected MCP integrations across NetSuite, Monday.com, HubSpot, Wrike, Google Ads, Adverity, Apify, Snowflake, LinkedIn, and Jasper.
Built the LinkedIn MCP OAuth flow through Auth0 and diagnosed the missing audience parameter that produced opaque tokens instead of JWTs.
Engineered the custom Jasper content-pack-pipeline server for Knowledge Base access, content brief generation, brand voice selection, and production workflows.
Produced a 34-page Monday.com MCP assessment covering feasibility, security posture, licensing economics, and rollout strategy.

Why it matters to hiring teams

Useful agents are mostly systems problems: authentication, tool boundaries, state, context, approvals, repeatability, observability, and workflow ownership. This platform shows the ability to turn messy enterprise systems into reusable AI infrastructure that non-engineering teams can actually use.

MCP Auth0 Claude SDK Claude Skills Jasper AI Snowflake HITL

Evidence

Agent work that changed operating leverage.

Platform scale

Enterprise SaaS behind Claude

Tooling covered finance, CRM, project management, advertising, Snowflake, LinkedIn, Jasper, content workflows, approvals, and reporting.

Reusable IP

23 Claude skills

Agency methodologies were packaged as reusable plugins instead of remaining tribal knowledge or loose prompts.

Workflow impact

Jasper content pipeline

A custom server connected Claude to Jasper Knowledge Base and brand voice selection for four enterprise B2B client workflows.

Need agents that plug into real workflows?

I build tool-using AI systems with reusable skills, governance, and product-grade operating boundaries.

Start a Conversation Back to Home