Ten projects, in order.

The secret to learning to code is not talent — it's ordering. Do the right things in the right sequence and each project feels like a small, achievable step instead of a cliff face. These ten are built exactly that way. Project one teaches a single file upload and one call to OpenAI; project ten combines everything into a working clinical documentation tool with several features behind one dashboard. Nothing new appears before you're ready for it, and every concept you meet gets used again and again.

Every project is deliberately medical. You learn faster building tools you'd genuinely reach for, and by the end you'll have a portfolio of practical, AI-powered web apps — plus, more importantly, an instinct for how the pieces fit together. If you haven't set up your phone yet, do the first-app walkthrough first; it installs everything these projects assume you have.

For every project below you get three things: the goal (what the finished app does), the skills it teaches, and — later, in section 05 — the exact prompt to paste into your AI assistant to build it. The difficulty stars (⭐) climb gently as you go:

⭐	Beginner — one main idea, gentle pace.
⭐⭐	Beginner-plus — combines a couple of earlier ideas.
⭐⭐⭐	Intermediate — more moving parts and messier input.
⭐⭐⭐⭐	Advanced-beginner — pulling several features into one app.

Do them strictly in order. The whole design rests on each project quietly reusing what the previous one taught you.

• 01 · Voice Transcription ⭐

  Goal: a webpage where you upload an audio recording and get back a text transcript, using OpenAI's Whisper model.

  You learn: the absolute backbone — Node.js and the Express web framework, handling file uploads, HTML forms, API keys, environment variables, and how a frontend and backend talk to each other. Every later project stands on this one.

• 02 · Medical Letter Summariser ⭐

  Goal: paste or upload a clinic letter or discharge summary and get a concise summary under clear headings — diagnosis, important findings, medications, follow-up, and outstanding tasks.

  You learn: your first real taste of prompt engineering — writing instructions that shape how the AI responds — plus reading text, sending longer inputs, and displaying an AI answer cleanly.

• 03 · Patient Information Generator ⭐

  Goal: enter a condition, a patient age and a reading level; the AI produces easy-to-understand advice, including warning signs that need urgent attention and when to seek review.

  You learn: collecting several inputs from one form at once, and formatting a response so it's genuinely readable for a patient rather than a clinician.

• 04 · Drug Information Assistant ⭐⭐

  Goal: type a medication name and receive its drug class, common indications, side effects, contraindications, monitoring and patient-counselling advice, laid out in tidy cards.

  You learn: form validation (checking the input makes sense before using it), sharper prompt writing, and formatting structured output — plus a loading animation so the page feels alive while it waits.

• 05 · Medical Quiz Generator ⭐⭐

  Goal: choose a specialty, a difficulty and a number of questions; the AI generates multiple-choice questions with four options each, the correct answer, and an explanation, shown one at a time with a final score.

  You learn: loops and lists — how to display many AI-generated items in sequence — and building a small dynamic, interactive interface.

• 06 · Referral Letter Generator ⭐⭐

  Goal: fill in a detailed form — patient history, examination, investigations, current medication, reason for referral — and get back a professionally written referral letter, with copy and download buttons.

  You learn: handling large forms, building reusable prompt templates, formatting long text, and adding copy-to-clipboard functionality.

• 07 · Consultation Notes Formatter ⭐⭐⭐

  Goal: paste your rough consultation notes and get back structured clinical documentation under headings — presenting complaint, history, examination, assessment, differential diagnosis and plan.

  You learn: working with large free-text input, cleaning up messy data, and reliably reshaping it into a consistent structure.

• 08 · Medical Translation Assistant ⭐⭐⭐

  Goal: enter English medical advice, pick a target language, and get a direct translation, a simplified patient-friendly version, and a back-translation into English so you can verify nothing was lost.

  You learn: dropdown menus, producing several distinct AI responses from one request, and handling language selection.

• 09 · Flashcard Generator ⭐⭐⭐

  Goal: upload lecture notes or paste text and get revision flashcards — a question, an answer and key facts — that you can page through with next, previous and shuffle buttons.

  You learn: reading uploaded documents, parsing longer text, and asking the AI for reliably structured output you can display card by card.

• 10 · Clinical Documentation Assistant ⭐⭐⭐⭐

  Goal: the finale — one application combining everything. Transcribe audio, summarise the consultation, generate referral letters, produce patient information and format notes, all behind a single dashboard with navigation between modules.

  You learn: multiple API routes (separate behind-the-scenes endpoints for each feature), navigation between screens, a clean folder structure, and modular programming — genuine full-stack development. This is the project that ties the whole path together.

Read down this table and you can see the design: every project introduces exactly one new idea, layered on top of everything before it.

1 · Transcription	File uploads, the Whisper API
2 · Summariser	Prompt engineering
3 · Patient info	Forms and user input
4 · Drug assistant	Better prompts and formatting
5 · Quiz	Lists and loops
6 · Referral	Large forms and templates
7 · Notes	Structured output
8 · Translation	Multiple responses and languages
9 · Flashcards	Document processing
10 · Clinical assistant	Combining multiple AI features

Two projects a week is a comfortable pace. Here's a six-week plan that lands you at a real full-stack app by the end:

Week 1	Set up the phone (Termux, Node.js, the assistant) + build Voice Transcription.
Week 2	Letter Summariser · Patient Information Generator.
Week 3	Drug Information Assistant · Medical Quiz Generator.
Week 4	Referral Letter Generator · Consultation Notes Formatter.
Week 5	Translation Assistant · Flashcard Generator.
Week 6	Clinical Documentation Assistant — then review, debug and showcase.

Each project starts from a single detailed prompt pasted into your AI assistant. The reason these prompts work is that they're specific — they name the stack (Express, Node.js), the inputs, the exact output structure, and the documentation to produce. Vague prompts get vague apps; specific prompts get something you can actually run. Here are two full examples so you can see the pattern.

Project 2 — Medical Letter Summariser:

Create a complete Node.js web application that summarises medical letters using the OpenAI Responses API. Build an Express.js backend and a responsive frontend. Allow users to paste or upload a text document. Ask OpenAI to summarise the letter under these headings: Diagnosis, Important Findings, Medications, Follow-up, Outstanding Tasks. Display the summary in a clean layout with a copy-to-clipboard button. Use dotenv for the API key, include comprehensive beginner-friendly comments, and generate a README.md with complete setup instructions.

Project 6 — Referral Letter Generator:

Create a complete Node.js web application that generates professional medical referral letters using OpenAI. Build a form containing: patient age, gender, presenting complaint, medical history, examination findings, investigations, current medication and reason for referral. The AI should produce a professionally written referral letter. Include a Copy button and a Download as Text button. Build using Express.js, store the API key in a .env file, comment throughout, and generate complete documentation.

Notice the shared skeleton: what to build → the stack → the inputs → the exact output structure → the buttons/behaviour → security (.env) → comments → documentation. Every one of the ten prompts follows that same shape — once you can see it, you can write your own.

Pasting a prompt and running the result teaches you very little. The habit that actually builds skill is a three-step loop I run on every project. It costs a few extra minutes and repays them many times over.

• 1 · Generate

  Paste the detailed prompt and let the assistant build the whole project — backend, frontend and documentation. Don't run it yet.

• 2 · Review

  Ask the assistant to check its own work before you trust it. A prompt that works well:

  Review the entire project as if you are a senior software engineer. Fix any bugs, improve the folder structure, improve security, optimise performance, add beginner-friendly comments, ensure all dependencies are included, update the README if necessary, and confirm the project can run successfully in Termux.

  The AI catches many of its own mistakes here, before they ever reach you.

• 3 · Explain

  This is where the real learning happens. Ask:

  Explain this project to me as if I have never programmed before. Go through every folder and every file, explain what it does, how it communicates with the rest of the application, and why it is needed. Do not change any code unless I ask you to.

  Read the explanation slowly. Over ten projects this is what turns you from someone who pastes prompts into someone who understands what they've built.

Generate builds it, Review hardens it, Explain teaches you. Skip the last two and you have software you don't understand; keep them and every project leaves you a little more capable than the last.

After the ten, the real exercise: design and build your own tool for a problem you actually hit in clinic. No template, no prompt handed to you — just the concepts you've absorbed, the Generate → Review → Explain habit, and the confidence to describe what you want. That's the moment the training wheels come off, and it's where genuine understanding replaces following instructions.

By this stage you'll have gone from running a single Node.js script to building complete AI-powered web apps with frontends, backends, API integration and documentation — a real foundation for going further. In my experience it's also the point at which doctors stop asking whether they can build with AI, and start asking what to build next.

These projects are for learning and prototyping. Anything touching real patient data needs proper information-governance, consent and clinical-safety review before it goes anywhere near practice — build to learn first, deploy responsibly later.