Cough

Podcast 2 — In conversation

Speaker A: I have had a cough for almost two weeks now. It is not serious, just persistent and very annoying.

Speaker B: Two weeks is worth checking out with a doctor. A cough that hangs around that long can have several causes — post-viral irritation, dry air, allergies.

Speaker A: That is true. I keep meaning to go. The interesting thing is I used cough up in a meeting yesterday and someone looked confused.

Speaker B: In the informal sense — reluctantly hand something over?

Speaker A: Yes. I said the client finally coughed up the signed contract. My colleague thought I was talking about a respiratory episode.

Speaker B: Context usually saves you, but it is a good reminder that phrasal verbs can be genuinely ambiguous out of context. Cough up meaning to produce something reluctantly is very British informal. In American English they might say cough it up or just fork over.

Speaker A: Is there a difference between cough and clear your throat?

Speaker B: Clearing your throat is gentler and more deliberate — it is the small, controlled sound you make before speaking or to signal attention. A cough is usually involuntary and more forceful. That said, a polite cough — the kind you do to get someone's notice — sits somewhere between the two.

Speaker A: And in medical language, are there specific types of cough?

Speaker B: Several. A dry cough produces no mucus and tends to feel scratchy. A productive cough brings up phlegm, which is why doctors ask about it. A barking cough sounds like a seal — it often indicates croup in children. A whooping cough is a specific bacterial infection with a very distinctive sound after the coughing fit.

Speaker A: I did not know cough had such a rich medical vocabulary attached to it.

Speaker B: It is one of those words that appears simple but unlocks an entire clinical domain the moment you look closely. In medicine, the character of a cough is a diagnostic tool in itself.

Podcast 3 — Prompt engineering

Speaker B: Cough is a fascinating word for prompt engineering because it sits at the intersection of medical applications, health data systems, and natural language symptom tracking. All three areas are in high demand for developers right now.

Speaker A: I had not thought about the health data angle specifically. What kinds of prompts does cough unlock?

Speaker B: Medical symptom trackers, patient intake forms, diagnostic support tools, and public health monitoring systems. The word is precise enough to anchor a whole domain of application logic. Let me walk through seven examples.

Speaker B: PROMPT_1: Build a React CoughLogger component for a personal health tracking app. The user selects cough type from a dropdown — dry, productive, barking, or whooping — rates severity from one to ten using a slider, and adds an optional note. On submit, POST the entry to /api/symptoms/cough with timestamp, type, severity, and note. Display the last seven days of cough entries in a compact timeline below the form. Use a clean clinical colour scheme with deep teal and white.

Speaker A: The dropdown values map directly to real medical terminology — that is immediately useful for a health app.

Speaker B: PROMPT_2: Design a PostgreSQL schema for a symptom tracking platform. Tables: patient (id, name, date_of_birth, email), cough_entry (id, patient_id, recorded_at, cough_type, severity, duration_seconds, note), diagnosis (id, patient_id, condition, confirmed_at, doctor_id), and doctor (id, name, specialisation, email). Add an index on cough_entry patient_id and recorded_at for fast time-series queries. Add a view recent_cough_activity showing all entries from the last 30 days with patient name joined.

Speaker B: The cough_type and severity columns make filtering by clinical criteria straightforward.

Speaker A: What about an API endpoint for logging cough data?

Speaker B: PROMPT_3: Write a Node.js Express endpoint POST /api/cough-log that accepts patient_id, cough_type, severity, duration_seconds, and an optional note. Validate that severity is between one and ten and cough_type is one of dry, productive, barking, or whooping. Insert the validated entry into a cough_entry table and, if severity is eight or above, also insert a high-severity alert into an alerts table with status pending. Return the new entry ID and, if triggered, the alert ID.

Speaker A: The automatic alert for high-severity coughs is a genuinely useful clinical feature. And for AI-assisted diagnosis?

Speaker B: PROMPT_4: Build a Vue 3 symptom checker page. The user answers five questions about their cough — type, duration in days, whether it worsens at night, whether it produces coloured mucus, and whether they have a fever. On submit, send the answers to the OpenAI chat completion API with a system prompt instructing the model to act as a triage assistant, list two to four possible causes in plain English, and recommend whether the user should see a GP within 24 hours, within a week, or can self-manage. Display the response in a formatted card with a clear disclaimer that this is not medical advice.

Speaker B: The disclaimer is essential — always instruct the model to include it when generating health-adjacent content.

Speaker A: That disclaimer instruction is a good prompt engineering habit. What about a full application?

Speaker B: PROMPT_5: Generate a full-stack Cough Monitor dashboard for a GP practice. React frontend, Flask backend, PostgreSQL. Features: a patient list showing each patient's most recent cough entry, a chart of average severity over the past 30 days per patient, a high-severity alerts panel listing all unresolved alerts with a Reviewed button that sets status to reviewed, and a search bar filtering patients by name or cough type. Style with a professional medical theme: white cards, deep navy background, teal accents.

Speaker A: A complete clinical monitoring tool from one domain word. What about public health monitoring at scale?

Speaker B: PROMPT_6: Design a Python data pipeline that reads cough_entry records from PostgreSQL every hour, aggregates total entries and average severity by postcode district, and writes the aggregated results to a cough_heatmap table. Build a Flask endpoint GET /api/heatmap that returns the latest aggregated data as GeoJSON, keyed by postcode district, for rendering in a Leaflet.js map. Include a last_updated timestamp in the response.

Speaker A: A geographic cough heatmap — that is essentially an epidemiological tool built from a single symptom table.

Speaker B: PROMPT_7: Create a vanilla JavaScript cough frequency counter widget. A large central button labelled Cough plays a soft click sound using the Web Audio API when pressed and increments a count displayed prominently below it. Persist the daily count in localStorage keyed by today's date. Show a seven-day bar chart using a canvas element, drawn with the Canvas 2D API, displaying daily cough counts. Include a Reset Today button that clears today's count after a confirmation dialog.

Speaker A: A self-contained offline tool that requires no server at all — the whole medical concept captured in a few hundred lines of vanilla JavaScript.

Speaker B: The right domain word pulls in the right architecture. Cough as the domain noun gives you the data model, the UI vocabulary, and the clinical logic all at once.