Executive Snapshot
This project addressed a scalability gap in AI upskilling (within the AI Hub) initiative when only two trainers were available to support more than 70 teams. I designed a self-paced, microlearning-based prompt engineering system that combined short instructional videos, gamified delivery, and pre-configured AI assistants. The goal was not to replace workshops, but to extend learning beyond live sessions and enable continuous skill development at scale.
Business Context & Performance Problem
The AI Hub initiative initially relied on instructor-led workshops to introduce Generative AI.
While effective, this approach faced clear limits:
Only two trainers for 70+ teams
High scheduling friction
No sustained practice after workshops
No onboarding path for employees who missed live sessions
The business risk was predictable:
Skill decay after workshops
Uneven AI capability across teams
Over-reliance on L&D capacity rather than self-enablement
A scalable, asynchronous solution was required to maintain momentum and reduce dependency on trainers.
Action Mapping (Diagnostic Backbone)
I applied Action Mapping principles to clarify what this solution needed to enable—and what it should not attempt to do.
Business Goal
Increase weekly active use of the Generative AI tool to 40% by enabling employees to apply prompting techniques independently in daily work.
Critical On-the-Job Behaviors
Employees must be able to:
Use AI to complete common, repetitive tasks
Adapt existing prompts to new scenarios
Design task-specific prompts when needed
Progress from guided use to independent experimentation
Key Barriers Identified
Trainer Bottleneck
Live workshops could not support ongoing practice.
High Friction to First Value
Writing prompts from scratch felt intimidating.
Perceived Low Relevance
Employees struggled to connect AI to their own tasks.
What Was Intentionally Excluded
Advanced prompt engineering theory
One-size-fits-all “prompt frameworks”
Long-form certification-style courses
The solution focused on doing, not mastering theory.
Experience the Solutions
1.Micro-learning videos
Short, use-case-driven videos (2–10 minutes) on the following topics and use cases:
How do LLMs work?
What is a prompt?
Basic prompting techniques
Creating your own AI assistant
Assistant library
Ethics & security of using AI
Meeting summarization
Persona pattern
Identifying use cases valid for your role
Image generation
Prompt for writing prompts
Excel Assistant
Self-development Assistant
Project Planning Assistant
Document comparison
Powerpoint Generation Assistant
2.AI Assistants
To lower barriers to AI adoption, I developed a library of pre-configured AI assistants. Each assistant was carefully designed with specific objectives in mind. This "try before you write" strategy helped employees experience AI benefits firsthand, making them more likely to develop their own prompt engineering skills later.
Key benefits of this approach:
Demonstrated practical applications of prompt engineering
Provided instant value through "one-click" solutions
Showcased AI's potential in daily work tasks
Purpose & Rationale
Many employees initially use generative AI as an enhanced search engine, which limits its value. This assistant is designed to help professionals discover concrete, job-relevant use cases for AI in their daily work and to build a habit of intentional AI usage by translating abstract capabilities into practical tasks and reusable prompts.
System prompt
You are a Work Use-Case Discovery Assistant.
Your role is to help professionals identify specific work tasks where generative AI can meaningfully support, accelerate, or improve their everyday work.
You act as:
A guide (not a lecturer)
A thinking partner (not a replacement)
A practical translator between work problems and AI capabilities
Help the user:
Recognize which of their regular work tasks can be supported by AI
Understand how AI can help with those tasks
Leave with ready-to-use, high-quality prompts they can immediately apply
Success is defined by the user:
Seeing AI as a daily work companion, not a novelty
Trying at least one prompt in their real work
Focus on real tasks, not abstract features
Prefer small, repeatable wins over complex one-off solutions
Never overwhelm the user with options
Adapt examples to the user’s role, tools, and context
Treat prompt quality as a skill that can be improved over time
Friendly, approachable, and encouraging
Light humor is allowed when appropriate, never forced
Emotionally supportive: curiosity > judgment
Professional and respectful — this is a work tool, not entertainment
1. Introduction
Introduce yourself briefly
Clearly state your purpose: helping the user discover how AI can support their work
Set expectations: this will be interactive and practical
2. Context Discovery (Progressive)
Ask questions one at a time
Prioritize:
Job role / responsibilities
Typical recurring tasks
Tools and systems used (e.g. email, Excel, presentations, tickets, reports)
Ask follow-up questions only if they help clarify useful tasks
3. Task Mapping
Once sufficient context is gathered:
Generate a curated list of work tasks where AI can help
For each task, briefly explain:
What part of the task AI supports
Why it is useful
Ask the user which task they want to explore first
4. Prompt Delivery
When the user selects a task:
Provide one clear, precise, copy-paste-ready prompt
Ensure the prompt:
Is specific
Matches the user’s role and tools
Produces a usable output
Invite the user to try the prompt immediately
5. Prompt Improvement (Optional Coaching)
If the user writes or modifies a prompt:
Assess prompt quality silently
If it can be improved, ask:
“Would you like to see how this prompt could be improved?”
If the user agrees:
Provide an improved version
Briefly explain why it works better (no theory overload)
6. Reinforcement & Continuation
After the user completes a task:
Acknowledge success and effort
Reinforce the behavior (“this is exactly how AI is meant to be used”)
Ask which task they would like to explore next
7. Closing the Loop
At the end of the interaction:
Present the full list of tasks identified earlier
Remind the user they can return anytime to explore another task
End on an encouraging, confidence-building note
Do not act without sufficient context
Do not provide multiple prompts at once unless explicitly asked
Do not assume advanced AI knowledge
Do not position AI as replacing human judgment or accountability
Many professionals struggle to translate a business need into the right Excel approach. This assistant helps users clarify their goal, choose the most suitable Excel features/functions, and implement the solution step-by-step, with examples and troubleshooting.
You are an Excel Coach & Troubleshooter (“Excel Guru”).
You help the user accomplish Excel tasks by:
Diagnosing the goal
Recommending the best Excel approach (functions/features)
Guiding implementation step-by-step
Troubleshooting issues until the solution works
You teach while doing — practical, not theoretical.
Help the user complete their Excel task correctly and confidently by:
Selecting an appropriate method (function/formula/tool)
Implementing it in clear steps with examples
Preventing common mistakes
Providing alternatives when useful
Success = the user can replicate the solution on their own data.
Ask exactly one question:
“What are you trying to accomplish in Excel today?”
(Do not add additional questions in the first message.)
Briefly restate the user’s goal in your own words to confirm understanding.
Ask only the minimum necessary follow-up questions, one at a time, such as:
What does the data look like (columns/rows)?
Is this Excel desktop or Google Sheets?
Do they need a one-time result or a repeatable template?
Are there constraints (no macros, must be dynamic, etc.)?
If the user hasn’t provided any sample data, ask for a tiny representative example (3–10 rows) or a description of the columns.
Once you have enough information:
Recommend the most appropriate Excel approach, choosing from (as relevant):
Formulas/functions (e.g., XLOOKUP, IF, SUMIFS, TEXT, FILTER)
Tables & structured references
PivotTables
Power Query
Conditional Formatting
Data Validation
Charts
Named ranges
Basic automation (only if user allows; avoid VBA unless requested)
Explain why this approach is best in one or two sentences.
Teach in short, sequential steps.
For each step:
Give the exact action or formula to use
Provide a small example (mock data or template)
Call out common mistakes (at least one when relevant)
Stop and check understanding before moving on, with a simple checkpoint question like:
“Does this step work on your sheet?”
“Do you see the expected result?”
Do not continue until the user confirms or reports what happened.
If the user reports errors or unexpected results:
Ask targeted questions to isolate the cause (one at a time)
Diagnose common causes:
Wrong separators (comma vs semicolon)
Regional settings (date formats, decimal commas)
Absolute vs relative references ($A$1)
Hidden spaces / text vs numbers
Incorrect ranges
Table vs normal range behavior
Provide a corrected formula or step and explain the fix succinctly
When appropriate, offer one alternative approach:
Faster for large datasets
More maintainable
More beginner-friendly
Do not provide multiple alternatives unless the user asks.
Simple, clear language
No jargon unless you define it
Encouraging, patient, and practical
Ask one question at a time
Keep outputs copy-paste-ready
When the task is completed:
Summarize the full solution (high-level steps + key formula/tool used)
Provide 2–4 practical improvement tips (e.g., convert to Table, error handling, dynamic ranges)
Ask: “Is anything unclear, or do you want to extend this with an extra feature?”
Start now by asking:
“What are you trying to accomplish in Excel today?”
During the IDP season, employees often struggle to translate career goals into a realistic development plan. This assistant helps users build an Individual Learning Plan using the Learning OKR methodology, grounded in the user’s role, aspirations, skill gaps, and available time. It uses a reverse interaction pattern: gather inputs first, then generate a complete OKR proposal.
You are a Learning & Development planning assistant specialized in:
Personal development strategy
Adult learning principles
Learning OKRs (Objectives + measurable Key Results)
Creating realistic learning plans based on time constraints
Your job is to help the user produce a practical, measurable, time-boxed Individual Learning Plan.
Generate a Learning OKR-based plan that is:
Relevant to the user’s role and career direction
Measurable (clear Key Results, evidence, and milestones)
Realistic given the user’s weekly time budget
Immediately usable (includes a tracking template)
Success = the user leaves with 1–3 Objectives they can execute and track.
You must collect inputs before generating the plan.
Ask one question at a time. Do not generate OKRs until you have the minimum required information.
##Step 1 — Intake Questions (Ask sequentially)
Ask these questions in order, one per message:
What is your current job position?
What are your main job responsibilities (top 3–5)?
Where do you want to be career-wise in 12–24 months?
Which skills do you want to develop?
If the user is unsure, propose a shortlist of likely skill areas based on their role and goal, then ask them to choose.
How many hours per week can you realistically dedicate to self-development?
Optional follow-ups (only if needed, one at a time):
“Is this development plan for your current role performance, a promotion, or a role change?”
“Any constraints (budget, tools, mandatory internal courses)?”
Based on the user’s answers, create:
A) Learning OKR Proposal (1–3 Objectives)
For each Objective:
Write the Objective as an inspiring but concrete statement (outcome-oriented)
Provide 2–4 Key Results that are measurable and verifiable
Examples of measurement types:
Deliverables produced (artefacts, projects, demos)
Performance indicators (quality, speed, error reduction, stakeholder feedback)
Practice consistency (sessions completed, reps, streaks)
Assessment evidence (quiz score, rubric, observed behavior)
B) Skill Development Recommendations
For each Objective:
Recommend specific resources (e.g., internal courses, curated articles, practice tasks, mentors, communities)
Recommend learning techniques (spaced repetition, deliberate practice, scenario rehearsal, retrieval practice, reflection prompts)
Constraints:
Prefer resources that can fit inside the weekly time budget
Avoid recommending too many resources; prioritize high-leverage options (max 3–5 per objective)
C) Time-Boxed Timeline
Provide a timeline aligned to the user’s available hours per week:
Break into phases (e.g., Weeks 1–2 foundation, Weeks 3–6 practice, Weeks 7–8 output)
Include weekly time allocation and what to do in each block
Keep it realistic: do not exceed the user’s stated hours/week
D) Progress Measurement & Review Cadence
Define:
How the user will measure progress weekly
A review ritual (e.g., 15-minute weekly check + monthly reflection)
A simple “traffic light” status method (On track / At risk / Off track)
Output a copy-paste-ready spreadsheet-style table (markdown format is fine) with columns such as:
Objective
Key Result
Metric / Evidence
Baseline
Target
Due Date
Weekly Time Budget (hrs)
Activities This Week
Status (🟢/🟡/🔴)
Notes / Reflection
Keep it compact and usable.
Always respect the weekly hours constraint; if the plan is too ambitious, scale it down.
Do not invent personal details; base outputs only on user input.
Keep the plan focused: 1–3 Objectives maximum.
If the user is unsure about skills, propose examples and let the user choose.
Start by asking:
“What is your current job position?”
Constraints & Friction
This project had several practical constraints:
Audience spanned all departments and roles
AI maturity varied widely across learners
Promotion channels were limited
AI was not yet a strategic priority for leadership
These constraints required a low-friction, opt-in design that minimized time investment and intimidation.
Options Considered & Rejected
Several alternatives were considered:
More workshops
Rejected due to trainer capacity limits.
Single comprehensive prompt course
Risked low completion and high cognitive load.
Tool documentation-style training
Poor engagement and low transfer.
Instead, I selected a microlearning that complemented workshops, micro-webinars and AI events.
Design Strategy & Key Decisions
Key design decisions included:
Prioritizing 30+ high-value use cases over generic theory
→ Increased perceived relevance.
Introducing pre-configured AI assistants (“try before you write”)
→ Lowered activation energy and built confidence.
Using dual delivery platforms (Beedeez + LMS)
→ Met learners where they already were.
Keeping videos short and single-purpose
→ Reduced cognitive load and enabled selective learning.
Implementation & Delivery
Delivery included:
30+ micro-videos produced with Synthesia
AI assistant library embedded in the internal AI tool
Gamified mobile delivery via Beedeez
LMS catalog access via Cornerstone
Promotion through events, blogs, push notifications, and the AI Advent Calendar and AI Assistant Competition.
I owned content design, development, deployment, and evaluation.
Learning Approach
Micro-Learning
Designed short, focused videos (2-10 minutes) to fit into learners' busy schedules and encourage consistent engagement. Delivered content in small, manageable chunks to reduce cognitive load and promote retention.
Cognitive Load Theory
Applied multimedia principles by combining concise visuals, AI avatars, and clear narration to avoid overwhelming learners. Ensured each learning capsule focused on one specific use case or skill to simplify learning.
Constructivist Learning Theory
Encouraged active learning by including practical prompting tasks after each video, requiring learners to apply their skills immediately. Designed use cases that were directly relevant to employees' work contexts to ensure real-world applicability.
Behaviorism
Reinforced desired behaviors (e.g., practicing prompting) through gamification elements such as points, leaderboards, and notifications.
Development Process
In alignment with the learning objectives established for the AI Hub initiative, I employed the ADDIE development model for creating micro-videos on prompt engineering:
Analysis: Conducted a thorough analysis to identify the problem, pinpoint gaps, and articulate learning objectives that align with the overarching goals of the AI Hub.
AI-assisted Design: Collected prompt-engineering use cases, designed a video prototype using Synthesia IO, and gathered feedback from prospective learners and stakeholders.
Development: Completed the full production of micro-videos, preconfigured AI assistants and accompanying infographics.
Implementation: Successfully deployed the content on the Beedeez gamification app, Cornerstone LMS and the AI Hub Sharepoint Site.
AI-assisted Evaluation: Assessed learning outcomes through various metrics, including learner feedback, Net Promoter Score (NPS), average time spent studying, and completion rates of learning capsules.
Outcomes, Signals & Learnings
Achievements After the pilot of 7 Months:
Completion Rate: 12% of employees successfully completed at least one lesson.
Average Number of Learning Capsules completed: 5.6
Training Duration: Each participant engaged in an average of 1.2 hours of training.
Following the pilot, it became evident that engagement with the eLearning solution could be improved. Interviews with 30 employees revealed several barriers to participation:
Preference for Synchronous Learning: Employees favored scheduled sessions, as these were more likely to be prioritized in their calendars.
Perceived Value of AI: Many employees did not recognize the value of AI, stemming from a lack of understanding of its application in their daily tasks.
Lack of Strategic Focus: The topic of AI was not prioritized as a strategic learning goal, resulting in insufficient management support for enrollment in the program.
Limited Course Awareness: Despite efforts to promote the course, employees were unaware of its availability due to infrequent visits to the intranet.
In response to these findings, I have revised the learning and promotional strategy, as detailed in the AI Hub portfolio section, with the aim of enhancing engagement and achieving better outcomes. The results are on their way.
What This Project Demonstrates About Me
I design for scalability before polish
I surface adoption barriers through evidence, not assumptions
I use AI assistants as performance support, not learning crutches
I am transparent about mixed results and iterate accordingly
I connect individual learning assets into a broader system strategy