Cholera

Podcast 2 — In conversation

Speaker A: I was reading about John Snow and the 1854 cholera outbreak in London. I knew he was important to epidemiology but I had not realised how radical his water theory was at the time.

Speaker B: Right. The mainstream view was miasma, the idea that disease spread through foul air. Snow was telling his colleagues that everything they believed was wrong, using only a hand-drawn map and interview notes rather than any laboratory proof.

Speaker A: And they removed the pump handle eventually.

Speaker B: They did, though the outbreak was already declining by then. What matters is Snow's method: using spatial data to identify a disease source. That is the foundation of modern epidemiology.

Speaker A: What is the difference between cholera and typhoid? Both are waterborne.

Speaker B: They share a common risk factor, inadequate sanitation, but they are very different diseases. Cholera causes sudden extreme fluid loss. Typhoid causes sustained fever and systemic illness. Confusing them would be a serious clinical mistake.

Speaker A: And oral rehydration therapy is genuinely that simple? Salt, sugar, and water?

Speaker B: The principle is elegantly simple. Cholera kills primarily through dehydration. Replace the fluid and electrolytes lost and most patients survive. The specific glucose and salt ratio matters because of how the intestinal sodium-glucose cotransporter works, but the basic recipe is on the back of a WHO sachet. It is one of the most cost-effective medical interventions ever developed.

Podcast 3 — Prompt engineering

Speaker B: Today we look at cholera in prompt engineering for public health and medical software. The word is highly specific and immediately scopes the AI toward epidemiology, outbreak management, and patient care systems.

Speaker A: That specificity is valuable. Writing cholera outbreak rather than disease outbreak gives the AI a concrete epidemiological domain with well-established data models and clinical protocols.

Speaker B: Exactly. The word signals waterborne transmission, dehydration as the mechanism, and a public health context. Let us start with a surveillance dashboard.

Speaker B: PROMPT 1: Build a public health surveillance dashboard for tracking waterborne disease outbreaks. Display a real-time choropleth map showing confirmed cholera case counts per district, fetched from a PostgreSQL database. Include 14-day trend lines per district, a threshold alert system that flags districts where weekly case rates exceed ten per 100,000 population, and a data export function. Stack: React, Leaflet.js, Node.js, Chart.js, PostgreSQL.

Speaker A: The choropleth map is the right primary view. Cholera spreads from point sources so spatial clustering is exactly the signal you want to surface first.

Speaker B: Precisely. Now a water quality monitoring system.

Speaker B: PROMPT 2: Design a water quality monitoring API for a municipal network of sensors. Each sensor logs pH, turbidity, and estimated bacterial load every fifteen minutes. When readings suggest cholera-risk contamination, trigger automated alerts to water authority engineers and the public health team via email and SMS. Persist all readings in a time-series database with ten-year retention. Stack: Node.js, InfluxDB, Nodemailer, Twilio, Express API.

Speaker A: InfluxDB is the right store for that sensor frequency. A relational database would struggle with the write volume at fifteen-minute intervals.

Speaker B: Exactly. Now a hospital patient management module.

Speaker B: PROMPT 3: Build a patient management module for an infectious disease ward handling cholera admissions. For each patient track hourly fluid intake and output, oral rehydration therapy compliance, antibiotic administration, and daily weight. Automatically flag patients whose hourly fluid output exceeds intake by more than one litre and alert the ward nurse dashboard in real time. Stack: React, Node.js, Socket.io, PostgreSQL.

Speaker A: The fluid balance threshold is very cholera-specific. That one litre per hour figure comes directly from the WHO clinical guidelines.

Speaker B: Encoding clinical knowledge in the prompt means the AI generates code with correct business logic rather than a generic tracker. Now a vaccination campaign system.

Speaker B: PROMPT 4: Create a vaccination records system for a regional cholera immunisation campaign. Log each recipient by name, date of birth, village, dose number, vaccine batch code, and administering health worker. Calculate coverage rates per village and alert campaign managers when any village falls below 80 percent of the target population for first-dose coverage. Generate PDF district reports. Stack: Next.js, PostgreSQL, Puppeteer.

Speaker A: The 80 percent threshold maps to the WHO herd immunity recommendation for oral cholera vaccine. That is well-specified.

Speaker B: Now a field investigation module for contact tracing.

Speaker B: PROMPT 5: Build a field investigation module for a cholera outbreak response system. When a case is registered, prompt the investigator to record the patient's drinking water source, food consumed in the past 48 hours, and household contacts. Identify water sources and food vendors shared across multiple cases, ranked by frequency. Visualise transmission clusters as a network graph. Stack: React, D3.js, Node.js, PostgreSQL.

Speaker A: The 48-hour food and water history maps to the cholera incubation period. That is clinically grounded.

Speaker B: The incubation period is two hours to five days, so 48 hours captures most events. Finally, a clinical knowledge base.

Speaker B: PROMPT 6: Build an AI-powered clinical decision support tool for health workers responding to cholera in low-resource settings. The system answers questions about oral rehydration therapy dosing by age and weight, antibiotic selection for severe cases, indications for intravenous fluids, and isolation procedures. Index the WHO 2017 cholera management guidelines as the primary knowledge source. Stack: Python, LangChain, ChromaDB, FastAPI, React chat interface.

Speaker A: Specifying the WHO guidelines as the source rather than general training data makes this a reliable clinical tool rather than a generic chatbot.

Speaker B: That is the critical distinction. Cholera as the domain keyword combined with an authoritative source constrains the outputs to clinically valid information. The word does real work in every one of these prompts.