Agentic AI workforce
Eight specialised AI roles orchestrated in Claude Cowork Projects across content, sales, finance, HR, IT and operations, each governed by a written mandate. Lets a two-person team run a six-function business.
Trainer Profile · Capstone Consulting
Agentic AI practitioner, data & analytics trainer, IBM Enterprise Design Thinking Practitioner. Seventeen years in the room.
arvinyeo@capstoneconsulting.com.sg linkedin.com/in/arvinyeo github.com/arvinnereus
Positioning
Arvin walks into a classroom having shipped that morning. A live eight-role AI workforce handling content, sales, finance, HR, IT and operations. A self-hosted full-stack agentic platform with realtime voice. A WhatsApp support agent running on a local model with its own retrieval knowledge base.
His learners don't watch a demo — they fork the real reference build and leave with their own workflow deployed. Underneath the engineering is a seventeen-year training career grounded in organisational psychology and quantitative method: a double degree from Murdoch University, ACLP certification from IAL Singapore, and a proprietary ROI methodology applied to every engagement.
That combination is the point. He teaches the decision layer and the build layer — how to frame a business question as a measurable analysis, and how to ship the agent that acts on the answer. He can hold the room with a C-suite sponsor and the terminal with their developers, in the same session.
Most AI trainers explain the technology. Arvin operates it in production, then teaches what actually broke.
Built, not briefed
Everything below is running or shipped. It doubles as the teaching material — learners see the real system, including the parts that failed first.
Eight specialised AI roles orchestrated in Claude Cowork Projects across content, sales, finance, HR, IT and operations, each governed by a written mandate. Lets a two-person team run a six-function business.
A self-hosted Flask and WebRTC application replacing roughly US$300 a month of SaaS — browser-to-model realtime voice, ephemeral-token security, nine integrated APIs, 23 of 23 unit tests green through every refactor. A multi-hour vendor TLS outage was routed around in ten seconds with zero downtime. This is the build Accelerator learners fork.
"Project Abigail" — a WhatsApp inquiry agent on Ollama qwen2.5:3b with an eight-document retrieval knowledge base, answering first-touch enquiries autonomously on owned infrastructure.
Capstone's own standard for agent mandates, human oversight boundaries, escalation rules and responsible-AI operating limits. Taught as core curriculum, not an appendix.
Hands-on GitHub Copilot and Claude Code training for engineering audiences — real implementation patterns for senior technical leads and SMEs, not a tool tour.
Open source · github.com/arvinnereus
Shipped applications and Claude skills, in the open, with commit history. Anyone evaluating him as an AI trainer can read the code before booking the session.
arvinnereus/baruch
A fully local, botless AI meeting notetaker for macOS — whisper.cpp transcription, speaker diarization and voice recognition, templated AI notes via Ollama. Zero cloud, zero tokens.
PythonLocal LLM
arvinnereus/capstone-crm
Business CRM — pipeline, contacts, finance and analytics — with an AI content-creation module. Next.js running on Cloudflare Workers.
TypeScriptEdge / Workers
arvinnereus/megaapp1
The full-stack agentic business platform described above — the live reference build Accelerator participants fork on the day.
Flask · WebRTCReference build
arvinnereus/wwad
"What Will Apple Do" — a Claude skill that critiques any product artifact through Apple's product doctrine and returns ranked, concrete cuts plus an optional simplified redesign.
ShellClaude skill
arvinnereus/enterprise-design-thinking
A Claude skill that facilitates IBM Enterprise Design Thinking — the principles, the Observe-Reflect-Make loop, the Keys (Hills, Playbacks, Sponsor Users) and the activity toolkit.
SkillIndependent
Signature programmes
Every applied programme carries a shipping standard: a working artefact running in the learner's own business by the end of the last session.
Walks through the production reference build, then each participant ships their own variant — an agent wired into a real business process, not a sandbox.
You leave withA functional agentic app and a live workflow in your business.
Prompt engineering, workflow mapping and responsible-AI governance, applied to the participant's own operation rather than a generic case study.
You leave withA working Claude Project and a 30-day implementation plan.
Turning a business question into a measurable analysis: framing the hypothesis, choosing the right test, reading significance honestly, and moving from descriptive reporting to inference and prediction. Regression, experiment design, KPI and dashboard architecture — taught for decision-makers, not statisticians.
You leave withAn analysis plan for a live business question, and the judgement to know when the numbers do not support the decision.
Delivered to 192 professionals across 50-plus organisations in 2025 with a five-star programme rating. Prompt engineering, workflow automation and generative AI in application.
You leave withA funded, nationally recognised AI competency and working automations.
"Your Journey into Advanced AI" — hands-on GitHub Copilot and Claude Code for technical audiences, from prompt fundamentals through to production-grade AI-assisted development.
You leave withA working AI-assisted development workflow in your own repository.
Built and accredited from scratch; 72 professionals trained across 20-plus organisations in live commerce strategy and consumer psychology.
You leave withA creator pipeline and a conversion-tested content approach.
Vision Leadership, Customer Service Excellence, Team Dynamics, Para-Counselling and The Great Marina Bay Challenge™ — delivered across 500-plus organisations.
You leave withMeasured behavioural change, evaluated against TIME™.
Custom programme architecture built on Enterprise Design Thinking loops, plus train-the-trainer for partner and internal L&D ecosystems.
You leave withA curriculum your own facilitators can run without him.
Methodology
Total Intervention and Measurement of Effectiveness — Capstone's Kirkpatrick-based ROI methodology, invented by Arvin and applied to every engagement, so training outcomes are reported as evidence rather than sentiment.
Hills, Playbacks and Sponsor Users embedded in curriculum design — the programme is co-created with the client's real users before it is delivered to them.
Statistical method carried over from research training and taught in its own right: hypothesis framing, experiment and A/B design, regression, and the honest reading of significance — so learners can tell a real effect from a convincing chart.
Kolb learning cycles and Mezirow transformative learning, with Belbin, MBTI, DISC and DOPE diagnostics used to shape group composition and coaching — grounded in a psychology and organisational-psychology double degree.
A hard standard on every applied programme: the session is not complete until each participant has something running. It is the difference between AI awareness and AI adoption.
Credentials
Technical & analytical stack
Claude Cowork · subagent orchestration · Claude API and the Anthropic SDK · Claude Code & GitHub Copilot · Ollama with RAG · WebRTC realtime voice · OpenAI Realtime · provider-abstraction and ephemeral-credential patterns · Flask · Cloudflare R2 · Make.com · statistical modelling, regression and experiment design · KPI and dashboard architecture
Selected clients
Engagement modes
Book a session
Tell him the team, the timeframe and the outcome you need. He will come back with a programme shape and what your people will have running by the end of it.
Email Arvin