Why I Built These
At IBM, I kept running into the same time-consuming tasks: summarizing meeting transcripts, building presentation outlines, creating data visualizations for partner reviews. So I built custom GPT prompts to automate them. Each one has structured inputs, detailed system instructions, and consistent output formatting so the results are usable right away, no editing needed.
Meeting Recap Generator
Takes a raw meeting transcript and turns it into a structured, professional recap ready to send to stakeholders.
How It Works
- Input: Paste a raw transcript from Teams, Zoom, Otter.ai, or manual notes
- Processing: The GPT identifies speakers, pulls out key topics, decisions, and action items
- Output: A formatted recap with meeting metadata, summary, discussion points, action items with owners and deadlines, and next steps
Design Choices
- Handles messy, unformatted transcripts with speaker identification heuristics
- Separates decisions made from items still under discussion
- Action items always include: task, owner, and deadline (or "TBD" if not stated)
- Professional tone suitable for executive-level distribution
Use case: Weekly partner check-ins with 30+ nonprofit organizations. This turned a 45-minute transcript review into a 2-minute GPT interaction.
AI Workforce Learning Deck Designer
Generates structured presentation outlines for CSR and workforce development topics, following IBM's communication standards.
How It Works
- Input: Topic, audience, and key data points or themes
- Processing: Structures content into a narrative arc with data-backed slides
- Output: Slide-by-slide outline with titles, bullet points, speaker notes, and visualization suggestions
Design Choices
- Follows a narrative structure: Context, Challenge, Solution, Impact, Call to Action
- Each slide includes speaker notes for the presenter
- Suggests data visualizations (bar charts, timelines, maps) where they make sense
- Adapts tone depending on whether the audience is internal, external partners, or executives
Use case: Quarterly CSR impact presentations for IBM SkillsBuild leadership. Turns raw metrics into a visual narrative.
IBM Slide Deck Builder with D3.js
Creates data-rich slide decks with interactive D3.js visualizations you can embed directly into web-based presentations.
How It Works
- Input: Dataset, presentation context, and target audience
- Processing: Analyzes data patterns and selects the right visualization types
- Output: Slide structure with D3.js code snippets for each visualization, plus narrative framing
Design Choices
- D3.js visualizations are self-contained and ready to embed
- Supports bar charts, line graphs, pie charts, geographic maps, and network diagrams
- Color schemes follow IBM's Carbon Design System guidelines
- Each visualization includes accessibility attributes (ARIA labels, color-blind safe palettes)
Use case: Data-driven partner review decks showing learner engagement, completion rates, and geographic distribution across IBM SkillsBuild's global network.
SkillsBuild Learning Plan Architect
Designs partner-ready learning plans using IBM SkillsBuild content, mapping modules, labs, and credentials into clear learner progressions.
How It Works
- Input: Topic, target audience (workforce adult learners or college students), and any specific SkillsBuild content to include
- Processing: Selects appropriate catalogue content, sequences it from quick wins to credential pathways, and structures it into two sections
- Output: A fully formatted learning plan with section titles, descriptions, activity labels, and mandatory/optional tags
Design Choices
- Uses a "you" voice throughout — never "learners" or "students" unless specified
- Automatically adds YouTube reminder text when a section includes external video
- Enforces strict source rules: no academia content for non-college audiences
- Outputs a consistent structure: Intro → Practice → Credential
Use case: Building IBM SkillsBuild learning plans for nonprofit workforce partners. Turns a topic brief into a structured, platform-ready plan in one interaction.
What They Have in Common
All three prompts follow the same engineering pattern:
System Instruction Layer
A detailed persona, constraints, and output format defined in the system prompt. This keeps behavior consistent no matter what the user types in.
Structured Input Schema
Each prompt expects a specific input format (transcript, topic brief, or dataset) with optional parameters for customization.
Output Templating
Outputs follow rigid templates with consistent sections, formatting, and metadata. You can use them right away without post-editing.