Prompt Engineering & AI Tools

Practical AI Curriculum · Beginner to AI-Powered Professional

A structured 14-week programme at Tech Minds Academy (Abuja & Live Online) open to all students regardless of technical background. Move from understanding how Large Language Models (LLMs) actually generate text to designing reliable prompt systems, automating real business workflows with n8n and Zapier, and integrating the OpenAI and Anthropic Claude APIs into production web applications.

14Weeks

5Stages

60+Lessons

9Projects

0Prerequisites

Download Official 2026 Curriculum PDF

---

Programme Overview

Core Topics: LLM Fundamentals · Zero-Shot & Few-Shot · Chain-of-Thought · Prompt Templates · ChatGPT & Claude · AI Automation · OpenAI & Anthropic APIs · Capstone AI System

Stage	Focus	Duration	Level	Prerequisite
Stage 1	Understanding AI & Large Language Models	2 weeks (Weeks 1–2)	Beginner	None
Stage 2	Core Prompting Techniques	3 weeks (Weeks 3–5)	Beginner	Stage 1
Stage 3	Advanced Prompting & Prompt Design	3 weeks (Weeks 6–8)	Intermediate	Stage 2
Stage 4	AI Tools Mastery	3 weeks (Weeks 9–11)	Intermediate	Stage 3
Stage 5	AI Workflows, API Integration & Capstone	3 weeks (Weeks 12–14)	Advanced	Stage 4

---

Curriculum Rationale — Key Decisions Explained

Question	Tech Minds Academy Pedagogical Decision
Why prompt engineering?	AI tools are now embedded in every industry — marketing, law, finance, healthcare, public administration, and software engineering. A professional who can direct AI models precisely, reproducibly, and safely is measurably more productive than ten who use AI casually. Prompt engineering delivers immediate impact on work quality from Week 1.
Why no coding prerequisite?	The first four stages require zero programming knowledge. Prompt engineering is fundamentally a clear communication, logic, and systems-thinking discipline before it is a software one. Stage 5 introduces REST API calls in JavaScript for students who want to go deeper — while ensuring non-coders can still deliver an impactful automated capstone system.
Why cover both automation and APIs?	Visual workflow platforms like n8n, Zapier, and Make empower operations leads, marketers, and founders to build end-to-end AI pipelines without writing backend infrastructure. Covering the OpenAI and Anthropic APIs alongside them teaches aspiring developers how AI powers real web apps — making the course valuable across every career track.
What comes after this course?	Graduates who want to engineer full-stack AI software products proceed directly to Agentic Software Engineering & MCP (16 Weeks) — covering autonomous AI agents, tool calling, RAG, custom Model Context Protocol (MCP) servers, and automated eval suites — or pair this course with JavaScript Fundamentals.

---

Stage 01 · Understanding AI & Large Language Models

Duration: 2 Weeks (Weeks 1–2) · Level: Beginner
How AI models work under the hood and why structured prompting is a foundational career skill.

Before writing production prompts, you must understand the engine you are steering. Stage 1 demystifies Large Language Models — how next-token prediction works, why models hallucinate, how context windows behave, and how to transform vague questions into reliable instructions.

Week 1: How AI Language Models Work (6 Lessons · 4 Hours)

1. 01 — What is an AI language model? A plain-English explanation: Moving past hype to understand probabilistic language modelling and next-token prediction.
2. 02 — How LLMs are trained — data, patterns, and prediction: Pre-training on massive text corpora, supervised fine-tuning (SFT), and alignment via RLHF.
3. 03 — Tokens — how AI reads and counts text: Sub-word tokenisation, why 1,000 tokens is roughly 750 English words, and how token counts govern API pricing and latency.
4. 04 — Temperature and creativity — deterministic vs creative outputs: Controlling sampling randomness from temperature: 0.0 (analytical extraction, code, JSON) to temperature: 0.8+ (brainstorming, brand copy).
5. 05 — Context windows — how much the model can remember at once: Understanding working memory limits, "lost-in-the-middle" attention degradation, and managing long conversations.
6. 06 — The difference between ChatGPT, Claude, Gemini, and Llama: Comparing frontier proprietary models (OpenAI GPT-4o, Anthropic Claude Sonnet/Opus, Google Gemini) with open-weights ecosystems (Meta Llama).

IMPORTANT

Week 1 Project — 3-Model Comparison Report: Test the exact same 5 benchmark prompts (summarisation, Nigerian market reasoning, creative copywriting, tabular extraction, and logic) across 3 distinct AI platforms (e.g., ChatGPT, Claude, and Gemini) and document an evidence-based comparison of accuracy, tone, formatting fidelity, and failure cases.

Week 2: The Anatomy of a Prompt (6 Lessons · 4 Hours)

1. 01 — What is a prompt? Input, instruction, and context: Separating the task instruction from the reference material the model operates on.
2. 02 — Why the same question gets different answers: Sampling variance, implicit ambiguity, hidden conversation state, and system instructions.
3. 03 — The four components of a well-structured prompt: Mastering the RCIO Framework — Role, Context, Instruction, and Output Format / Constraints.
4. 04 — System prompts vs user prompts — roles and instructions: How persistent system-level rules govern persona, safety boundaries, and behaviour across multi-turn chats.
5. 05 — What AI cannot do — hallucinations, knowledge cutoffs, limits: Recognising fabricated citations, arithmetic traps, and temporal cutoffs, plus grounding techniques to prevent them.
6. 06 — Evaluating AI output — how to spot a bad response: Building a verification checklist for factual grounding, completeness, tone drift, and subtle logical flaws.

Practical Example: Weak Prompt vs Structured RCIO Prompt

❌ WEAK PROMPT (Vague, ungrounded, unpredictable format)
Write an email to customers saying our delivery in Abuja will be late today.

✅ STRUCTURED RCIO PROMPT (Role · Context · Instruction · Output Format)
[ROLE]
You are the Customer Experience Lead for SwiftBasket, a same-day grocery delivery service operating across Abuja (Wuse, Maitama, Gwarinpa, and Bwari).

[CONTEXT]
Heavy afternoon rainfall and traffic congestion along the Kubwa Expressway have delayed afternoon dispatch riders by 90–120 minutes. All chilled items remain stored in insulated cold boxes. Customers can reply to reschedule for tomorrow morning at zero extra fee.

[INSTRUCTION]
Write a proactive customer notification message informing affected customers of the delay, reassuring them about product freshness, and offering the free morning reschedule option.

[OUTPUT FORMAT & CONSTRAINTS]
- Provide two versions:
  1. Email (clear subject line + 120–150 words body)
  2. WhatsApp broadcast message (max 65 words, warm and concise)
- Tone: Empathetic, accountable, calm, Nigerian professional English.
- Do NOT use clichés like "We sincerely apologise for any inconvenience caused."

IMPORTANT

Week 2 Project — Rewriting 5 Weak Prompts: Take 5 vague, real-world prompts across business email, research, customer support, data extraction, and learning; rewrite each using the 4-part RCIO structure; and document side-by-side Before/After outputs explaining why each structural change improved reliability.

---

Stage 02 · Core Prompting Techniques

Duration: 3 Weeks (Weeks 3–5) · Level: Beginner · Prerequisite: Stage 1
The foundational, research-backed techniques every competent AI practitioner reaches for automatically.

Stage 2 equips you with the core toolkit proven to transform erratic model responses into repeatable, structured outputs: few-shot exemplars, Chain-of-Thought reasoning, negative constraints, and strict JSON generation.

Week 3: Zero-Shot, Few-Shot & Role Prompting (6 Lessons · 5 Hours)

1. 01 — Zero-shot prompting — asking without examples: When direct instructions suffice and where zero-shot prompts break down on custom formatting or nuanced style.
2. 02 — Few-shot prompting — teaching by example: Supplying input-to-output demonstrations so the model learns exact structure, length, and voice in-context.
3. 03 — How many examples are enough? Finding the right number: Balancing 2-shot, 3-shot, and 5-shot exemplars against token cost and overfitting to example quirks.
4. 04 — Role prompting — assigning a persona to the AI: Specifying domain expertise, audience calibration, and decision-making priorities rather than superficial titles.
5. 05 — Combining role + few-shot for highly specific outputs: Pairing an expert persona with gold-standard examples to lock in both analytical depth and exact brand voice.
6. 06 — When each technique works best — practical decision guide: Choosing between zero-shot, role-only, and few-shot prompting for speed vs consistency.

Copy-Pasteable Few-Shot E-Commerce Prompt Block

You are a Senior E-Commerce Copywriter for an electronics & solar retailer serving Nigerian households and SMEs.
Convert raw product specs into a consistent listing matching the exact style, structure, and Naira formatting of the examples below.

### Example 1
Input: Felicity 5kVA Hybrid Inverter | 48V | MPPT 100A | Price: 920000 | Warranty: 2 years
Output:
**Felicity 5kVA 48V Hybrid Solar Inverter (100A MPPT)**
Keep your home office, freezer, and lighting running seamlessly through grid outages. Built with an integrated 100A MPPT solar charge controller, this 48V hybrid inverter switches automatically in milliseconds so your Wi-Fi and workstations never reboot.
- **Price:** ₦920,000
- **Best For:** 3-bedroom homes, clinics, and retail shops
- **Peace of Mind:** Backed by a 2-year local warranty

### Example 2
Input: Starlink Standard Actuated Kit | High-speed satellite internet | 150Mbps+ | Price: 590000 | Warranty: 1 year
Output:
**Starlink Standard High-Speed Satellite Internet Kit**
Stream, deploy code, and host crystal-clear client calls from anywhere in Nigeria — from central Abuja to remote project sites. Delivers low-latency broadband speeds over 150 Mbps within minutes of plug-and-play setup.
- **Price:** ₦590,000
- **Best For:** Remote engineers, agencies, and multi-user offices
- **Peace of Mind:** Backed by a 1-year hardware warranty

### Now Generate Listing
Input: EcoFlow River 2 Pro Portable Power Station | 768Wh LiFePO4 | 70-min fast charge | Price: 645000 | Warranty: 5 years
Output:

IMPORTANT

Week 3 Project — E-Commerce Few-Shot Generator: Build and test a reusable Role + Few-Shot prompt that transforms messy supplier notes into consistent, conversion-ready product descriptions across 10 store items without manual reformatting.

Week 4: Chain-of-Thought & Structured Reasoning (6 Lessons · 5 Hours)

1. 01 — Chain-of-thought prompting — making AI show its work: Why generating intermediate reasoning tokens dramatically improves accuracy on math, logic, pricing, and strategy tasks.
2. 02 — Step-by-step instructions — breaking tasks into stages: Writing explicit numbered reasoning phases so the model analyses evidence before committing to a conclusion.
3. 03 — Self-consistency — asking the model to verify its own answer: Prompting independent reasoning checks and critique passes to catch arithmetic and assumption errors.
4. 04 — Tree-of-thought — exploring multiple reasoning paths: Instructing the model to generate 3 distinct strategic options, evaluate trade-offs for each branch, and converge on the strongest path.
5. 05 — When to use reasoning prompts vs direct prompts: Avoiding unnecessary latency and verbosity on simple lookups while reserving reasoning scaffolds for multi-constraint problems.
6. 06 — Applying chain-of-thought to complex real-world tasks: Structuring financial unit-economics checks, vendor evaluations, and market entry analyses.

IMPORTANT

Week 4 Project — Multi-Step Business SWOT Chain: Design a single structured reasoning prompt that conducts market sizing, a grounded SWOT analysis, risk scoring, and a 90-day strategic recommendation for a real local or digital business model.

Week 5: Formatting, Constraints & Output Control (6 Lessons · 5 Hours)

1. 01 — Specifying output format — JSON, markdown, tables, bullet points: Using explicit schema definitions and visual skeletons to guarantee machine- and human-readable layouts.
2. 02 — Length control — asking for short, medium, or detailed responses: Enforcing word ceilings, sentence counts, and paragraph budgets accurately.
3. 03 — Negative constraints — telling the model what NOT to do: Eliminating filler intros ("Certainly! Here is..."), banned buzzwords, markdown wrappers, and unsupported claims.
4. 04 — Tone and style instructions — formal, casual, technical, friendly: Defining concrete voice attributes (reading level, vocabulary, sentence rhythm) instead of vague adjectives.
5. 05 — Asking for multiple options — generating variants for comparison: Generating A/B/C headline, hook, and positioning variants across distinct angles in a single pass.
6. 06 — Structured JSON output — using AI as a data generation tool: Enforcing valid JSON arrays, data types, enum values, and null handling for downstream spreadsheets and web apps.

Copy-Pasteable Structured JSON Product Data Generator

Generate realistic seed catalogue data for a Nigerian bookstore & edtech shop as a raw JSON array of 10 items.

Strict Rules:
1. Return ONLY valid JSON — no markdown code fences, no preamble, and no trailing commentary.
2. Every object MUST follow this exact schema:
   - "id": string formatted as "PROD-001" through "PROD-010"
   - "name": string (max 60 characters)
   - "price_ngn": integer (between 4500 and 85000, multiples of 500)
   - "in_stock": boolean
   - "category": one of ["Programming", "Business", "Exam Prep", "Hardware Accessories"]
   - "description": string (20 to 35 words, plain text, no emojis)
   - "tags": array of exactly 3 lowercase strings

IMPORTANT

Week 5 Project — Structured JSON Product Data Generator: Write and validate a deterministic prompt that outputs a clean JSON array of 10 items (name, price, description, tags, category) that parses cleanly in JSON.parse() with zero syntax edits across 3 consecutive test runs.

---

Stage 03 · Advanced Prompting & Prompt Design

Duration: 3 Weeks (Weeks 6–8) · Level: Intermediate · Prerequisite: Stage 2
Moving from one-off chat messages to reusable, debuggable, multi-step prompt systems.

Professionals do not rewrite prompts from scratch every morning. Stage 3 teaches you how to parameterise prompts with clean variables, systematically debug failing prompts using changelogs, and chain multiple prompts into reliable pipelines.

Week 6: Prompt Templates & Reusability (6 Lessons · 5 Hours)

1. 01 — What is a prompt template? Variables and placeholders: Separating fixed system logic from dynamic runtime inputs.
2. 02 — Designing templates for repeatable tasks: Identifying recurring team workflows (SOP drafting, ticket triage, meeting synthesis, code review) worth templating.
3. 03 — Building a content generation template — blog, social, email: Creating modular templates with clear input blocks and output contracts.
4. 04 — Parameterising prompts — {topic}, {tone}, {audience}: Designing unambiguous variable placeholders ({topic}, {tone}, {audience}, {call_to_action}) with default fallbacks.
5. 05 — Storing and organising your prompt library: Structuring personal and team prompt repositories in Notion, Markdown/GitHub, or Raycast/TextExpander snippets.
6. 06 — Sharing and versioning prompts — prompt management tools: Documenting usage instructions, sample inputs, expected outputs, and model compatibility for teammates.

Parameterised Prompt Template Example

# Template ID: TM-CONTENT-04 (B2B Educational Newsletter)
# Recommended Model: Claude 3.5/3.7 Sonnet or GPT-4o | Temperature: 0.4

SYSTEM / INSTRUCTIONS:
You are an Industry Newsletter Editor writing for {audience}.
Draft an educational newsletter issue on the topic below.

INPUT PARAMETERS:
- Topic: {topic}
- Tone: {tone}
- Primary Pain Point: {pain_point}
- Call to Action: {cta_offer}

STRUCTURE REQUIREMENTS:
1. 3 Subject Line options (under 48 characters each)
2. Opening hook anchored in {pain_point} (max 60 words)
3. 3 actionable lessons with concrete Nigerian/African workplace examples
4. One-paragraph sign-off leading naturally into {cta_offer}

CONSTRAINTS:
- Total body length: 350–450 words.
- Avoid generic AI transitions ("In today's fast-paced digital landscape", "Delve into").

IMPORTANT

Week 6 Project — 5-Template Prompt Library: Build a documented, reusable library of 5 parameterised prompt templates covering content creation, code explanation/generation, deep research synthesis, stakeholder email communication, and data/business analysis.

Week 7: Iterative Refinement & Prompt Debugging (6 Lessons · 5 Hours)

1. 01 — Why first-draft prompts rarely work perfectly: Treating prompt design as engineering iteration rather than guesswork.
2. 02 — The prompt refinement loop — test, evaluate, adjust, repeat: Building a 5-input edge-case test set (short input, messy input, adversarial/ambiguous input) before judging prompt quality.
3. 03 — Common failure modes — vague, over-constrained, conflicting: Spotting contradictory rules, buried instructions, and cognitive overload in monolithic prompts.
4. 04 — Diagnosing a bad prompt — what went wrong and why: Isolating whether an error stems from missing context, ambiguous schema definitions, wrong model tier, or temperature settings.
5. 05 — A/B testing prompts — comparing two versions systematically: Scoring Prompt Variant A vs Variant B blind across rubric criteria (accuracy, format compliance, conciseness).
6. 06 — Keeping a prompt changelog — tracking what worked: Maintaining semantic version history (v1.0 → v1.1 → v2.0) so improvements are never lost.

IMPORTANT

Week 7 Project — 3-Iteration Prompt Debug Log: Take a brittle, failing real-world prompt through 3 documented refinement iterations (v1.0 baseline → v1.1 constraint fix → v1.2 edge-case hardening → v2.0 production release), recording test inputs, failure diagnoses, rubric scores, and changelog notes at each step.

Week 8: Meta-Prompting & Prompt Chaining (6 Lessons · 6 Hours)

1. 01 — Meta-prompting — asking the AI to write or improve your prompt: Using frontier reasoning models as prompt architects and rubric critics to sharpen your own templates.
2. 02 — Prompt chaining — feeding output of one prompt into the next: Why decomposing complex jobs into focused sequential prompts beats giant "do-everything" mega-prompts.
3. 03 — Building a multi-step workflow with chained prompts: Passing clean XML/JSON handoff artefacts (<research_brief>, <outline>, <draft>) from Step 1 to Step 4.
4. 04 — Conditional chaining — different chains based on AI output: Designing classification routers (e.g., routing a customer message to Billing, Technical Support, or VIP Escalation chains).
5. 05 — Avoiding context loss across long chains: Passing distilled state summaries alongside raw outputs so final steps stay faithful to original user constraints.
6. 06 — Designing a prompt pipeline for a real business task: Blueprinting inputs, intermediate schemas, quality gates, and human review checkpoints.

IMPORTANT

Week 8 Project — 4-Step Research-to-Article Chain: Engineer a linked 4-step prompt pipeline (Step 1: Research Topic & Fact Extraction → Step 2: Structured Argument Outline → Step 3: Full Long-Form Draft → Step 4: Editorial & Tone Polish) with clean structured handoffs between every stage.

---

Stage 04 · AI Tools Mastery

Duration: 3 Weeks (Weeks 9–11) · Level: Intermediate · Prerequisite: Stage 3
Using leading AI platforms productively across text, cited research, spreadsheets, coding, and visual media.

Prompt engineering without deep platform fluency is theory without leverage. Stage 4 trains you on the flagship ecosystems shaping modern workplaces: ChatGPT Custom GPTs, Claude Projects, Perplexity, NotebookLM, GitHub Copilot, Cursor, and Midjourney / Ideogram.

Week 9: ChatGPT & Claude — Text & Reasoning (6 Lessons · 5 Hours)

1. 01 — ChatGPT in depth — GPT-4o, custom GPTs, memory, plugins: Navigating OpenAI model tiers, Data Analysis sandboxes, memory controls, and Custom Instructions.
2. 02 — Claude in depth — long context, document analysis, coding: Leveraging Claude's 200K+ token context window, Artifacts preview environment, and nuanced writing fidelity.
3. 03 — Choosing the right model for the task at hand: Matching speed, reasoning depth, vision capability, and context size to each daily task.
4. 04 — Uploading documents — analysing PDFs, spreadsheets, images: Auditing contracts, financial statements, CSV exports, and whiteboard screenshots safely and accurately.
5. 05 — Building custom GPTs — instructions, knowledge, and actions: Packaging system instructions, uploaded knowledge-base files, conversation starters, and guardrails into shareable assistants.
6. 06 — Claude Projects — persistent context for ongoing work: Organising client workspaces, brand style guides, and technical specs into persistent Claude Project knowledge bases.

IMPORTANT

Week 9 Project — Custom GPT for a Nigerian Small Business: Design, configure, and stress-test a domain-specific Custom GPT (or Claude Project) for a Nigerian SME — such as an Abuja logistics firm, solar installer, pharmacy, or fashion label — grounded in a custom FAQ document, Naira product/service catalogue, escalation rules, and brand tone guidelines.

Week 10: AI for Content, Research & Productivity (6 Lessons · 5 Hours)

1. 01 — AI-assisted writing — blog posts, newsletters, social media: Human-in-the-loop drafting workflows that preserve authentic expertise and local context while eliminating blank-page friction.
2. 02 — Research workflows — summarising papers and reports with AI: Extracting methodology, key statistics, limitations, and executive takeaways from dense policy or industry PDFs.
3. 03 — Perplexity AI — AI-powered search with citations: Using live web search, Focus modes, and source verification to eliminate hallucinated facts during market and competitor research.
4. 04 — NotebookLM — deep research on your own documents: Grounding Q&A, briefing docs, timelines, and audio overviews strictly inside your uploaded source notebooks with inline citations.
5. 05 — AI for spreadsheets — formulas, data cleaning, analysis: Generating complex Excel/Google Sheets formulas (XLOOKUP, INDEX/MATCH, QUERY, regex cleaning) and pivot summaries via natural language.
6. 06 — Building a weekly content calendar workflow with AI: Connecting audience research, pillar themes, multi-channel repurposing, and publishing schedules.

IMPORTANT

Week 10 Project — 4-Week Social Content Calendar: Execute an end-to-end research-to-publishing workflow using Perplexity/NotebookLM + Claude/ChatGPT to research an industry niche, define content pillars, draft 20+ platform-native posts, and deliver a structured 4-week content calendar spreadsheet.

Week 11: AI Coding Tools & Image Generation (6 Lessons · 5 Hours)

1. 01 — GitHub Copilot — AI code completion in VS Code: Setting up VS Code + GitHub Copilot for inline completions, comment-driven scaffolding, and chat explanations.
2. 02 — Claude and ChatGPT for code — debugging, explaining, refactoring: Reading stack traces, explaining unfamiliar scripts line by line, and refactoring HTML/CSS/JavaScript cleanly.
3. 03 — Cursor IDE — AI-native code editor: Multi-file context (@Files, @Docs), Composer/Agent edits, and reviewing diffs responsibly before accepting AI-generated code.
4. 04 — AI image generation: Midjourney, DALL-E, and Ideogram: Strengths of each model — photorealism and editorial aesthetics in Midjourney, conversational edits in GPT-4o/DALL-E, and crisp typography in Ideogram.
5. 05 — Writing effective image prompts — style, composition, lighting: Mastering the Visual Prompt Formula: [Subject & Action] + [Environment] + [Composition & Camera Angle] + [Lighting & Colour Palette] + [Medium/Style] + [Aspect Ratio Parameters].
6. 06 — AI video tools: Runway, Sora, and Kling — an overview: Understanding image-to-video motion prompts, camera movement controls, storyboard consistency, and commercial ethics.

IMPORTANT

Week 11 Project — JS Utility + 10 Consistent Brand Assets: Part A: Use GitHub Copilot / Cursor to build and test a functional browser JavaScript utility (e.g., an invoice/VAT calculator or WhatsApp link builder). Part B: Use Midjourney or Ideogram to produce a cohesive 10-asset visual brand pack maintaining consistent colour palette, lighting, and visual identity.

---

Stage 05 · Building AI-Powered Workflows & Capstone

Duration: 3 Weeks (Weeks 12–14) · Level: Advanced · Prerequisite: Stage 4
Automating business operations end-to-end and integrating the OpenAI and Anthropic APIs into live web applications.

The final stage bridges prompt engineering with software automation. You will wire event-driven workflows in n8n, Zapier, and Make, write JavaScript fetch() integrations with streaming and multi-turn conversation history, and ship a documented Capstone AI System.

Week 12: AI Workflow Automation (6 Lessons · 6 Hours)

1. 01 — What is AI automation? Triggers, actions, and AI steps: Designing event-driven pipelines (Trigger → Data Enrichment → LLM Reasoning Node → Router → Downstream Action).
2. 02 — n8n — open-source workflow automation with AI nodes: Running self-hosted or cloud n8n workflows using webhooks, HTTP requests, structured output parsers, and AI Agent nodes.
3. 03 — Zapier AI — connecting apps and adding AI processing steps: Connecting Google Forms, Gmail, Slack, Notion, and Airtable with OpenAI/Claude steps in minutes.
4. 04 — Make (Integromat) — visual automation with AI integration: Multi-branch routers, iterators, JSON aggregators, and error-handling fallbacks in visual scenarios.
5. 05 — Building a lead qualification workflow with AI scoring: Scoring inbound sales enquiries (0–100), tagging urgency/budget tier, and alerting sales reps automatically.
6. 06 — Building an AI-powered email response workflow: Drafting context-aware replies from a company knowledge base with human-approval drafts in Gmail/Outlook.

IMPORTANT

Week 12 Project — Automated Intake-to-Report Workflow: Build a live n8n, Zapier, or Make automation where a new form submission → AI classifies intent, urgency, and sentiment + summarises key details → sends a formatted HTML email report and logs a structured row to Google Sheets/Airtable.

Week 13: Using the OpenAI & Anthropic APIs (6 Lessons · 7 Hours)

1. 01 — What is an AI API? When to use it vs a chat interface: Understanding HTTP endpoints, JSON payloads, API keys, rate limits, and why production apps never hardcode secret keys in public frontend code.
2. 02 — The OpenAI API — authentication, models, and the messages array: Structuring developer/system, user, and assistant message roles, temperature, and response_format.
3. 03 — The Anthropic API — Claude models, system prompts, and max_tokens: Working with https://api.anthropic.com/v1/messages, top-level system parameters, and Anthropic version headers.
4. 04 — Calling the API from JavaScript with fetch(): Writing clean async/await HTTP requests, handling non-200 status codes, and parsing token usage metadata.
5. 05 — Streaming responses — displaying output as it generates: Reading Server-Sent Events (ReadableStream / TextDecoder) so users see words appear in real time instead of staring at a frozen spinner.
6. 06 — Building a simple AI chatbot in the browser with vanilla JS: Managing the conversation history array across turns, rendering clean UI bubbles, and wiring a local Node/edge proxy for API key safety.

Runnable JavaScript fetch() Examples — OpenAI & Anthropic Claude APIs

Below are clean ES2024+ fetch() implementations demonstrating multi-turn conversation history and real-time Server-Sent Events (SSE) streaming (runnable in Node.js 20+ with OPENAI_API_KEY / ANTHROPIC_API_KEY set in your environment, or via a lightweight backend route for browser apps):

// 1. Calling OpenAI Chat Completions API with Conversation History & Streaming
const conversationHistory = [
  {
    role: 'system',
    content:
      'You are Tech Minds Career Advisor in Abuja. Give concise, practical guidance in clear Nigerian professional English.'
  }
]

export async function askOpenAIStream(userMessage, onChunk) {
  conversationHistory.push({ role: 'user', content: userMessage })

  const response = await fetch('https://api.openai.com/v1/chat/completions', {
    method: 'POST',
    headers: {
      'Content-Type': 'application/json',
      Authorization: `Bearer ${process.env.OPENAI_API_KEY}`
    },
    body: JSON.stringify({
      model: 'gpt-4o-mini',
      temperature: 0.3,
      stream: true,
      messages: conversationHistory
    })
  })

  if (!response.ok) {
    const errorText = await response.text()
    throw new Error(`OpenAI API Error (${response.status}): ${errorText}`)
  }

  const reader = response.body.getReader()
  const decoder = new TextDecoder('utf-8')
  let fullReply = ''

  while (true) {
    const { done, value } = await reader.read()
    if (done) break

    const chunk = decoder.decode(value, { stream: true })
    for (const line of chunk.split('\n')) {
      if (!line.startsWith('data: ') || line === 'data: [DONE]') continue
      const parsed = JSON.parse(line.slice(6))
      const token = parsed.choices?.[0]?.delta?.content ?? ''
      if (token) {
        fullReply += token
        onChunk(token)
      }
    }
  }

  conversationHistory.push({ role: 'assistant', content: fullReply })
  return fullReply
}

// 2. Calling Anthropic Claude Messages API with Top-Level System Prompt
export async function askClaude(userMessage, history = []) {
  const response = await fetch('https://api.anthropic.com/v1/messages', {
    method: 'POST',
    headers: {
      'Content-Type': 'application/json',
      'x-api-key': process.env.ANTHROPIC_API_KEY,
      'anthropic-version': '2023-06-01'
    },
    body: JSON.stringify({
      model: 'claude-sonnet-4-20250514',
      max_tokens: 1024,
      temperature: 0.2,
      system:
        'You are a Senior SME Operations Analyst. Always structure recommendations with clear Naira (₦) cost assumptions and bulleted action items.',
      messages: [...history, { role: 'user', content: userMessage }]
    })
  })

  if (!response.ok) {
    throw new Error(`Anthropic API Error (${response.status}): ${await response.text()}`)
  }

  const data = await response.json()
  return {
    reply: data.content[0].text,
    usage: data.usage // { input_tokens, output_tokens }
  }
}

IMPORTANT

Week 13 Project — Branded Streaming AI Assistant in the Browser: Build a custom-branded web assistant featuring a domain-specific system prompt, multi-turn conversation memory, real-time token streaming, conversation reset, and safe API key handling via a lightweight local proxy.

Week 14: Capstone Project (6 Sessions · 10 Hours)

1. 01 — Defining the problem — what task will AI solve? Scoping a high-value workflow with measurable success criteria and clear boundaries.
2. 02 — Designing the prompt system — templates, chains, constraints: Architecting system instructions, few-shot exemplars, JSON schemas, and fallback rules.
3. 03 — Building the interface or workflow — web app or automation: Implementing either a clean browser application or a multi-node n8n/Zapier automation system.
4. 04 — Connecting the AI API — system prompt, user input, response: Wiring OpenAI or Anthropic endpoints, structured parsing, and error recovery.
5. 05 — Testing and refining — evaluate outputs across varied inputs: Running at least 15 normal, messy, and adversarial test cases and documenting refinements.
6. 06 — Documentation — prompt rationale, workflow diagram, usage guide: Publishing a complete portfolio repository and live demo walkthrough.

IMPORTANT

Week 14 Capstone AI System: Deliver a complete, tested, and documented AI-powered system. Choose one of four industry tracks (or propose an approved custom business problem):

1. Customer Support Triage & Resolution Bot (grounded in company policies with human handoff)
2. Multi-Channel Brand Content Generator (research → angle → platform-ready assets)
3. CV & Job-Fit Analyser (structured rubric scoring, gap analysis, and tailored interview prep)
4. SME Business Intelligence Tool (turns raw weekly sales/feedback CSVs or notes into executive action reports)

---

Assessment & Completion

Every student is evaluated through practical execution rather than rote memorisation:

Component	Description	Weight
Weekly Projects (×9)	Prompt library, comparison study, JSON generator, Custom GPT, workflow automation, or API app built and documented across the stages	40%
Stage Assessments (×5)	Practical prompt-engineering live challenge and written conceptual evaluation at the end of each stage	30%
Capstone AI System	Complete AI-powered web tool or automated workflow — architected, stress-tested across varied inputs, and documented	30%

---

Learning Outcomes — What You Can Do on Completion

On graduating from Prompt Engineering & AI Tools, you will be able to:

• Explain clearly how Large Language Models work, including tokenisation, context windows, and temperature sampling.
• Write structured, reliable prompts using RCIO, zero-shot, few-shot, role, and Chain-of-Thought techniques.
• Control AI output format (strict JSON, Markdown tables), tone, length, and negative constraints with precision.
• Design reusable, parameterised prompt templates ({topic}, {tone}, {audience}) and maintain a versioned prompt library.
• Diagnose, A/B test, and iteratively debug brittle prompts using documented prompt changelogs.
• Build multi-step prompt chains and automated business workflows using n8n, Zapier, or Make.
• Use ChatGPT, Claude, Gemini, Perplexity, and NotebookLM productively across research, spreadsheets, and content operations.
• Generate cohesive brand imagery and visual assets using Midjourney, DALL-E, and Ideogram, and accelerate coding with GitHub Copilot and Cursor.
• Call the OpenAI and Anthropic Claude APIs from JavaScript using fetch(), including streaming responses and conversation history.
• Design, build, evaluate, and document a complete AI-powered capstone system ready for your professional portfolio.

---

Tools & Resources — Everything You Need

Tool	Purpose in the Curriculum	Cost Tier
ChatGPT (OpenAI)	Primary AI assistant — GPT-4o, Data Analysis, and Custom GPT creation	Free / Plus
Claude (Anthropic)	Long-context AI — deep document analysis, coding, Artifacts, and Claude Projects	Free / Pro
Perplexity AI & NotebookLM	Cited real-time web research and source-grounded notebook synthesis	Free
Midjourney / DALL-E / Ideogram	AI image generation for consistent brand, marketing, and creative assets	Paid / Free tiers
n8n / Zapier / Make	Open-source and cloud visual AI workflow automation platforms	Free (self-host n8n) / Free tiers
VS Code + GitHub Copilot / Cursor	Code editing and AI-assisted JavaScript utility & API development	Free tier
OpenAI API	Programmatic REST access to GPT models for web applications and automations	Pay-as-you-go
Anthropic API	Programmatic REST access to Claude models for structured apps and workflows	Pay-as-you-go

---

Next Steps in Your Tech Minds Pathway

• Ready to build autonomous multi-step AI software, RAG pipelines, and custom MCP servers? Continue directly to Agentic Software Engineering & MCP (16 Weeks).
• Want deeper frontend & backend web engineering foundations alongside AI? Explore JavaScript Fundamentals (14 Weeks) and Node.js & Backend Development (15 Weeks).
• Setting up your laptop for Stage 5? Follow the Developer Environment Setup Guide and Git & GitHub Workflow Guide.