DELOITTE INNOVATION HUB · AI UPSKILLING PROGRAM

Take control of AI. Think like an engineer. Build with confidence.

255 Developers · 8 Weeks · 6 Cohorts · Deloitte Innovation Hub Egypt

255
Developers
8
Weeks
6
Cohorts
5
Facilitators
PROGRAM PHILOSOPHY

Ctrl+AI is the Deloitte Innovation Hub's internal AI upskilling program for 255 developers across frontend, backend, mobile, UI/UX, and testing. Rather than teaching specific tools, the program instills a software engineer mindset for the AI era — universal principles that work regardless of discipline, tool, or model. Role-specific application comes in Week 7. Everything before that is shared, intentional, and immediately applicable.

FORMAT

Half day per week (3–4 hrs). Live sessions on Teams. Sandbox tasks on Azure. Content archived on Notion.

REGISTRATION

First-come first-served. 40–50 devs per cohort. 6 rolling cohorts over ~38 weeks to cover all 255.

TOOLS COVERED

Claude, GitHub Copilot, Cursor, Figma AI, Stitch, MCP servers, VS Code, Android Studio, Xcode, Playwright AI.

FACILITATOR TEAM

Martin + 5 Leads

Facilitator Role Owns Weeks
Sorsok, Martin
Program Lead		 LEAD Overall design, Cohort 1, leadership reporting, co-facilitates all core sessions. Engineer mindset module. All
Abouzaid, Hagar
UI/UX		 UI/UX UI/UX track — Figma AI, Stitch by Google, design-to-code workflows, prompt-to-wireframe, model selection by design role W2, W7
Moheb, Mohamed
Frontend		 FRONTEND Frontend track — MCP → React pipeline, code review agents, UI governance & accessibility, component generation W3, W4, W7
Saleeb, Sandra
Mobile		 MOBILE Mobile & engineering practices — Spec-Driven Dev, SKILL.md, Claude code review, Prompt Engineering deep dive W3, W4, W5, W7
Saweres, Kirollos
Mobile		 MOBILE Foundations + DevOps — ML vs LLM, Token optimization, Automation & CI/CD, Offline Models, domain-specific IDE setup W1, W4, W6, W7
Sherif, Youssef
Backend		 BACKEND Backend track — Claude Code deep dive, MCP servers, Plan mode, Subagents, Unit tests, Mermaid generation W2, W3, W5, W7
FULL CURRICULUM

8-Week Program

Core (W1–W6) — All disciplines together Role Labs (W7) — Split by discipline Capstone (W8) — Demo Day
W1
Foundations — ML, LLM, Why AI, and the Engineer Mindset
3.5 hrs · Live session · No sandbox yet · Facilitators: Saweres, Kirollos + Martin
CORE THEORY MINDSET
+
SESSION TOPICS
ML vs LLM — What's the difference and why it matters
Saweres, Kirollos
Traditional ML (pattern recognition, training loops) vs. Large Language Models (next token prediction, emergent reasoning). Why developers need to understand both to work effectively with AI tools.
Why Using AI — Pros & Cons — Honest assessment for engineers
Saweres, Kirollos
Where AI genuinely speeds up developer work. Where it creates false confidence. How to build the habit of critical review. The "junior engineer" mental model.
The Engineer Mindset — Input → Context → Output
Martin Sorsok
The universal frame for all 8 weeks. Every technique taught in this program optimizes one of these three levers. Most devs only think about Input. The best AI users obsess over Context.
Help on My Daily Work — AI mapped to your real workflow
Saweres, Kirollos + Martin
Live demo: map AI to every stage of the dev cycle — planning, coding, review, testing, documentation. Where it saves the most time per discipline.
☁ SANDBOX ACTIVITY
No sandbox yet. Activity: Open poll — "What do you think AI is?" Capture assumptions. Each dev writes one sentence in Notion: "I currently use AI for ___." Revisit at Week 8.
W2
Prompt Engineering + Claude Code Introduction
4 hrs · Live workshop · ☁ Sandbox · Facilitators: Sherif, Youssef + Moheb, Mohamed
CORE PROMPTING CLAUDE CODE
+
SESSION TOPICS
Effective Prompt — Structure, role, context, constraints, format
Sherif, Youssef + Saweres, Kirollos
The anatomy of a good prompt. Role assignment, context injection, task specification, output format, constraints. Chain-of-thought for complex engineering tasks. When to be precise vs. when to be open-ended.
Claude Code Introduction — Overview, commands, tools & hooks
Sherif, Youssef
What Claude Code is and how it differs from Claude chat. Core commands, slash commands, context management. Tools and hooks — how to extend Claude Code's behavior. Setup on each dev's machine.
Figma AI + Stitch by Google — Design-to-code with AI
Moheb, Mohamed
How to generate Figma designs using Stitch by Google and Figma AI. Prompt-to-wireframe workflows. When to use these tools and when not to. Targeted at UI/UX devs but demonstrated for all.
Prompt Engineering — Deep dive and practice
Saleeb, Sandra
Iterative prompting workflow. Before/after comparisons. The most common prompting mistakes developers make. How prompt quality directly affects output quality — live demonstration.
☁ SANDBOX ACTIVITY
Take a real bug from your current project. Solve it using Claude Code with documented prompt iterations. Compare: vague prompt vs. structured prompt — screenshot both outputs. Post to #ctrl-ai-showcase on Teams.
W3
MCPs, Agents, and Spec-Driven Development
4 hrs · Live workshop · ☁ Sandbox · Facilitators: Sherif, Youssef + Saleeb, Sandra
CORE MCP AGENTS SPEC-DRIVEN
+
SESSION TOPICS
What is MCP + How to Work With It — Model Context Protocol explained
Sherif, Youssef + Saweres, Kirollos
What MCP servers are and why they matter. How they extend Claude's reach into your tools (GitHub, Figma, databases, APIs). Setting up and connecting MCP servers in Claude Code. Live demo: connecting an MCP server to a real project.
MCPs → React Components — Turning Figma designs into code
Moheb, Mohamed
How to use MCP servers to transform Figma designs directly into React components. The full pipeline: Figma → MCP → Claude → production-ready component code.
Plugins, Skills, Subagents, Tips & Tricks
Sherif, Youssef
Claude Code plugins ecosystem. SKILL.md files — what they are and how to create them for your project. Subagents for parallel tasks. Practical tips that save time daily.
Spec-Driven Development — Let AI build from specifications
Saleeb, Sandra
Writing specs that AI can execute reliably. The difference between a vague brief and a machine-readable spec. How Spec-Driven Development reduces hallucination and increases output quality. Live demo: spec → implementation.
☁ SANDBOX ACTIVITY
Write a SKILL.md file for your current project or team. Connect one MCP server to your sandbox environment. Use it to complete a small real task. Document what worked and what the MCP unlocked that you couldn't do before.
W4
Code Generation, Review, and Token Optimization
4 hrs · Live workshop · ☁ Sandbox · Facilitators: Saleeb, Sandra + Saweres, Kirollos
CORE CODE REVIEW COST
+
SESSION TOPICS
Claude Code Review Feature — How to use it well
Saleeb, Sandra
Using Claude as a code reviewer — not just a generator. Prompting for bugs, security issues, performance, and style. How to give Claude enough context for a useful review. Before/after examples with real diffs.
How to Create AI Agent for Code Review
Moheb, Mohamed
Building an automated code review agent using Claude. How to set it up to run on PRs. What to automate vs. what needs human eyes. Integration patterns for existing CI workflows.
How to Optimize Tokens and Cost
Saweres, Kirollos
Why token costs matter even with a budget. Context window management. Caching strategies. Batch processing. How to write prompts that are both effective and efficient. Cost estimation for production AI workflows.
Plan Mode and Specs — Before you generate, plan
Sherif, Youssef
Claude Code's plan mode — how to use it to architect a solution before generating code. Writing machine-readable specs inside plan mode. Iterating on the plan before committing to implementation.
☁ SANDBOX ACTIVITY
Take a piece of legacy code from your team. Run it through Claude's code review. Then refactor it using plan mode first, then implementation. Submit: original code, review output, plan, and final refactor as a before/after report in Notion.
W5
Testing, Debugging, and Documentation with AI
3.5 hrs · Live workshop · ☁ Sandbox · Facilitators: Sherif, Youssef + Saleeb, Sandra
CORE TESTING DEBUGGING DOCS
+
SESSION TOPICS
Unit Tests with AI — Generate, review, and trust
Sherif, Youssef
AI-generated unit tests — what to trust, what to always check. Edge case coverage. Test-first prompting: describe behavior, let AI generate the test, then implement. Using Claude Code's subagents for parallel test generation.
UI Governance, Guardrails & Accessibility Testing
Moheb, Mohamed
Using AI to enforce design system compliance. Automated accessibility audits. Protecting UI code from regressions with AI-assisted unit tests. How to apply guardrails by role and function in your codebase.
Mermaid Diagram Generation — Document with AI automatically
Sherif, Youssef
Using Claude to generate Mermaid diagrams from code — architecture, flow, sequence, ERD. Keeping documentation in sync with code changes automatically. Embedding diagrams in Notion and markdown docs.
MD Files and Documentation Diagrams
Saweres, Kirollos
AI-powered markdown documentation. Auto-generating README files, API docs, and architecture docs from code. How to maintain living documentation that stays accurate as code evolves.
☁ SANDBOX ACTIVITY
Pick a real function or module from your codebase. Use AI to: (1) generate unit tests, (2) generate a Mermaid architecture diagram, (3) write a README section. Compare quality vs. what you'd write manually. Time each task.
W6
Security, Deloitte Policy, Automation & Staying Current
3.5 hrs · Live session · Facilitators: Saweres, Kirollos + Sorsok, Martin
CORE SECURITY POLICY CICD
+
SESSION TOPICS
Client Data Security & Deloitte Policy — What you must know
Sorsok, Martin
What client data can and cannot be sent to AI models. Deloitte Innovation Hub's approved AI tools and sandboxed environments. How to work with AI tools without breaching client confidentiality. Real examples of what goes wrong and how to avoid it.
Offline Models — AI without internet or data risk
Saweres, Kirollos + Sorsok, Martin
When and why to use offline/local models (Ollama, LM Studio). Trade-offs vs. cloud models. Use cases where offline models are the right call for Deloitte client projects. Setup and practical usage.
Automation & CI/CD with AI
Saweres, Kirollos
Integrating AI into your pipeline: automated code review on PRs, AI-generated changelogs, test generation in CI, documentation auto-update on merge. How to build AI-augmented DevOps without introducing fragility.
How to Stay Updated on AI Improvements
Sorsok, Martin + Saweres, Kirollos
The AI landscape moves weekly. How to build a personal system for staying current without information overload. Recommended sources, how to evaluate new models and tools, and the monthly AI Drop channel on Teams.
☁ SANDBOX ACTIVITY
Policy quiz: Given 5 real scenarios (involving client data, external APIs, public models), decide: safe to use AI or not? Then: add one AI step to your team's CI pipeline — even a simple one. Document what you added and what it does.
W7
Role Labs — Same Mindset, Your Context
4 hrs · Split by discipline · ☁ Full sandbox day · All facilitators
ROLE LABS FRONTEND BACKEND MOBILE UI/UX TESTING
+
🖥️ FRONTEND LAB — Moheb, Mohamed
Figma AI → React Component full pipeline using MCP
Full workflow: Stitch by Google → Figma AI design → MCP server → Claude Code → production React component. Participants build a complete UI feature end-to-end.
UI Governance & Accessibility — applying guardrails in code
AI-assisted accessibility audit on a real component. How to enforce design system rules using Claude. Automated accessibility unit tests.
⚙️ BACKEND LAB — Sherif, Youssef + Martin
Spec-Driven API development — spec → tests → implementation
Write an OpenAPI spec, generate tests from it using AI, then implement the endpoint using Claude Code plan mode. Full TDD loop with AI assistance.
Subagents for parallel backend tasks — documentation + tests simultaneously
Using Claude Code subagents to generate API docs and unit tests in parallel. Integration with CI pipeline.
📱 MOBILE LAB — Sorsok, Martin
Domain-specific AI setup — VS Code, Android Studio, Xcode
Copilot in Android Studio. Claude Code in VS Code for Flutter. GitHub Copilot in Xcode. How to configure each IDE's AI tools for maximum productivity on mobile.
AI for cross-platform feature scaffolding — Flutter/RN patterns
Building a full screen with AI assistance in your mobile IDE of choice. Widget generation, state management setup, and navigation scaffolding.
🎨 UI/UX LAB — Moheb, Mohamed
Model selection by role/function — Which AI for what design task
Comparison of models for design work: Claude for copy & reasoning, Midjourney for visuals, Figma AI for layout, Stitch for prototypes. Decision framework for each use case.
Research synthesis & copy generation — AI in the design process
Using Claude to synthesize user research into themes. AI-generated microcopy, error messages, and onboarding text. Prompt-to-wireframe full workflow.
🧪 TESTING LAB — Saleeb, Sandra + Saweres, Kirollos
AI test case generation from specs and user stories
Given a user story, generate a complete test plan using AI. Then generate the test code. Review for coverage gaps. Automation with Playwright AI.
Regression automation & bug triage with AI
AI-powered root cause analysis for failed tests. How to use Claude to interpret error logs and suggest fixes. Building a self-healing test suite pattern.
☁ SANDBOX ACTIVITY — BUILD A COMPLETE FEATURE IN YOUR ROLE
Each discipline builds a small but complete feature using the tools from their lab. This becomes the foundation of the Week 8 capstone. Document every AI tool used, every prompt that worked, and what you'd do differently.
W8
Capstone + Demo Day — Ctrl+AI Certification
4 hrs · Demo Day · All facilitators + Leadership invited · Deloitte Innovation Hub
CAPSTONE CERTIFICATE LEADERSHIP
+
DEMO DAY FORMAT
5-Minute Presentation Per Developer
Each developer presents: (1) The problem they solved, (2) The AI tools and workflow used, (3) One prompt that surprised them, (4) What they'd do differently. Peer + facilitator scoring against shared rubric.
Certificate Ceremony — Foundation, Practitioner, or Champion
Certificates awarded on the day based on attendance, capstone quality, and peer contribution. Champions announced — they co-facilitate the next cohort.
Revisit Day 1 — "I currently use AI for ___"
Pull up each developer's Week 1 Notion entry. Compare to what they actually built. This is the program's most powerful closing moment — the before/after is always striking.
NPS Survey + Next Cohort Invitation
Collect NPS on the session. Champions invited to help shape the next cohort. Top projects featured in leadership quarterly report. Community channel stays active — this isn't the end.
🏆 CAPSTONE DELIVERABLE
A real deliverable from your actual work — built using AI tools from this program. Not a toy project. Not a tutorial. Something you'd be comfortable showing to your team lead. Document the workflow, not just the output.
CERTIFICATION

Ctrl+AI Certificate — 3 Levels

Issued by Deloitte Innovation Hub Egypt. LinkedIn-shareable. Feeds into performance review cycle. Champions teach the next cohort.

🥉
AI Foundation
REQUIREMENT: Attend 6+ of 8 sessions
Understands the engineer mindset for working with AI. Knows the mental model: Input → Context → Output. Can apply AI tools to daily developer tasks.
🥈
AI Practitioner
REQUIREMENT: Foundation + pass capstone Demo Day
Demonstrated applied AI workflow in a real production context. Can use Claude Code, MCP servers, prompt engineering, and role-specific tools end-to-end.
🥇
AI Champion
REQUIREMENT: Practitioner + co-facilitate a session in the next cohort
Certified to train and mentor peers. Listed in Deloitte Innovation Hub leadership quarterly report. Eligible to lead role labs in future cohorts.
CERT VALUE TO THE DEVELOPER
📄  Appears on Deloitte Innovation Hub internal profile and performance review
🔗  LinkedIn-shareable — "Ctrl+AI by Deloitte Innovation Hub" branding
🏅  Champions listed by name in leadership quarterly report
🎤  Champions co-facilitate next cohort — real visibility, real career move
♻️  Annual re-certification cycle keeps the credential meaningful
ROLLOUT PLAN

6 Rolling Cohorts

COHORT 1 — PILOT

Martin runs personally. All 5 facilitators co-present and observe. ~45 developers. Feedback collected session-by-session. Program iterated before Cohort 2.

COHORTS 2–6

Facilitators rotate lead ownership. Martin as program lead and escalation. Champions from Cohort 1 invited to co-facilitate. Community self-sustains on Teams.

CohortPeriodDevsNote
Cohort 1Week 1–8~45PILOT — Martin leads
Cohort 2Week 7–14~50Facilitators rotate
Cohort 3Week 13–20~50Champions co-facilitate
Cohort 4Week 19–26~50Champions co-facilitate
Cohort 5Week 25–32~40Full program maturity
Cohort 6Week 31–38~20Final wrap
DELOITTE INNOVATION HUB

AI Core — The Attendee Sandbox

🧠
WHAT IS AI CORE?

AI Core is the Deloitte Innovation Hub's internal AI environment — a governed, secure sandbox that gives program attendees direct access to AI tools, models, and APIs to build, experiment, and run demos without using personal accounts or breaching client data policy.

Every developer enrolled in Ctrl+AI gets access to AI Core for the duration of their 8-week cohort. All sandbox tasks, live demos, and capstone projects are run through AI Core — keeping everything within Deloitte's approved environment.

🔒

Secure & Policy-Compliant

All AI interactions happen within Deloitte's governed environment. No client data leaves the sandbox. Fully aligned with the Deloitte Innovation Hub's data and security policies covered in Week 6.

⚡

Ready from Day One

No setup required from attendees. AI Core is pre-configured with access to Claude, GitHub Copilot, and approved APIs. Developers log in with their Deloitte credentials and start building immediately from Week 2.

🏗️

Built for Real Projects

AI Core isn't a toy environment. Attendees use it to solve real problems from their actual work. The capstone on Demo Day is built entirely inside AI Core — making it a genuine showcase of what's possible within Deloitte's infrastructure.

WHAT ATTENDEES CAN DO IN AI CORE
→
Run all weekly sandbox tasks
Each week's hands-on activity is executed directly in AI Core — no personal API keys or accounts needed.
→
Experiment with AI models (Claude, Copilot, and more)
Prompt different models, compare outputs, and understand their strengths — all within a safe, pre-approved environment.
→
Build and submit the Week 8 capstone project
The Demo Day deliverable lives in AI Core — reviewable by facilitators and shareable with leadership as a verified internal project.
→
Access MCP server demos and integrations
Week 3's MCP session runs live demos directly in AI Core — no external configuration, no access issues during the session.
→
Try offline/local models safely
Week 6 covers offline models for client-sensitive work — AI Core provides a pre-configured environment to test these without IT friction.
AI CORE IN THE PROGRAM — WEEK BY WEEK
W1No sandbox — foundation concepts only. AI Core access granted at end of session.
W2First AI Core task — prompt a real bug, compare structured vs. vague prompts in Claude.
W3Connect an MCP server inside AI Core. Build a SKILL.md. Run a Figma → React pipeline.
W4Code review, plan mode, and token optimization tasks — all in AI Core with Claude Code.
W5Generate unit tests and Mermaid diagrams from a real module in AI Core.
W6Security policy quiz + CI/CD integration task + offline model trial in AI Core.
W7Full role lab day — each discipline builds a complete feature in AI Core using their stack.
W8Capstone Demo Day — all projects submitted and presented from AI Core.
PROGRAM INVESTMENT

Budget & Cost Breakdown

Total monthly budget: $250/month — covering the facilitator team subscriptions and API access for demo and content preparation.

CONFIRMED · $150 / MONTH
Claude Team Plan
Standard tier · 6 seats · $25/seat/month
6 SEATS ALLOCATED
MS
Sorsok, Martin
Program Lead · Full access
LEAD
HA
Abouzaid, Hagar
UI/UX · Figma AI, Design workflows
UI/UX
MM
Moheb, Mohamed
Frontend · MCP → React, Copilot
FRONTEND
SS
Saleeb, Sandra
Mobile · Spec-Driven, Claude Code review
MOBILE
SK
Saweres, Kirollos
Mobile · ML/LLM, Token optimization, CI/CD
MOBILE
SY
Sherif, Youssef
Backend · Claude Code, MCP, Subagents
BACKEND
WHAT THE TEAM PLAN INCLUDES
✓  Access to Claude Sonnet 4.6 & Opus 4.6
✓  Higher usage limits than Pro (more than enough for demo prep)
✓  Shared Projects — facilitators collaborate on session content
✓  Central billing — one invoice, Deloitte finance-friendly
✓  Admin controls — Martin manages all 6 seats
✓  Works on web, desktop, iOS, Android
OPTIONAL · UP TO $100 / MONTH
Claude API Key
For demo preparation & content that needs programmatic access
WHY THE API KEY?

The Team Plan gives conversational Claude access. But some session demos require programmatic API calls — building live MCP integrations, running automated code review agents, or showing Batch API token cost optimization. The API key covers these cases.

WHEN YOU'LL NEED IT
W3 MCP server live demo — requires API calls to connect tools
W4 Automated code review agent — runs as a script, not in chat
W4 Token & cost demo — Batch API comparison needs real API calls
W5 CI/CD integration demo — AI step in pipeline needs API key
ALL Facilitator content prep — generating exercises, rubrics, slide content
COST ESTIMATE (PAY-AS-YOU-GO)
Sonnet 4.6 — session demos~$20–30
Content & exercise generation~$20–30
MCP / agent demo scripts~$10–20
Buffer for Cohort 1 iteration~$20
Estimated total$70–100 / month

💡 Tip: Use Batch API for all non-live content generation — it's 50% cheaper than standard API calls. Only use real-time API for live session demos.

MONTHLY BUDGET SUMMARY
Claude Team Plan — Standard
6 seats × $25/seat/month
$150
Claude API Key — Pay-as-you-go
Demo prep, agents, CI/CD integration, content generation
≤$100
Total Monthly Budget
Attendees use AI Core — no additional cost per developer
$250/mo

Key point: The $250/month covers the 6-person facilitator team only. The 255 developers attending each cohort use AI Core provided by Deloitte Innovation Hub — there is no per-attendee subscription cost for the program.