Dr. Rakesh Kumar
CHART 02 / SYSTEM ARCHITECTURE

System Architecture

One patient, six modules, one record — how a visit moves from registration through to a final diagnosis, and which module owns each step.

6 Modules Patient Journey

Why name it this way

Every project on this site so far has been built as its own standalone tool. In practice they're stations on the same patient journey, each owning one part of the record before handing the patient — and the data — to the next. Naming them as modules makes that handoff explicit, and gives the two investigation modules that don't exist yet (Histopathology, Radiology) a defined place to slot into later.

The patient journey

Module 1

Registration Module

Patient registers at the counter or online.

Built — Project 1
Module 2

Vitals IoT Module

Demographics, chief complaints, and vitals captured at the kiosk — printed on the registration slip.

Proposal — Project 2
patient goes to OPD / Emergency
Module 3

Doctor's Exam Module

Clinical history and examination, as in the department template. Ends in Provisional Diagnosis 1.

Built — Project 3
investigations ordered
Module 4

Blood INV Module

Biochemistry, Haematology, and Pathology (blood-based). Results feed into Provisional Diagnosis 2 — which can differ from Diagnosis 1.

Data modelled — Project 4
Module 5

Histopathology Module

Tissue / biopsy specimens, reported separately from blood investigations.

Not yet built
doctor re-consulted with results; radiology investigations ordered
Module 6

Radiology Module

Imaging reported independently, reaching its own Radiological Diagnosis.

Not yet built

Combined Report

Radiological Diagnosis + Diagnosis after blood investigations, brought together.

Decision point

Final Diagnosis

Made by the doctor, reconciling all three inputs — clinical, blood/tissue, and radiological.

Doctor's call

The point of splitting it up

The diagnosis isn't a single event — it's revised at least twice: once from history and examination alone, once after blood (and eventually tissue) results come back, and once more when radiology is combined in. Keeping each module's output as its own labelled field, rather than overwriting a single "diagnosis" box, is what makes it possible to later ask questions like how often does the final diagnosis differ from the provisional one — which is exactly the kind of question a structured, module-tagged record (like the sample dataset) is built to answer.

Why every module is a structured checklist, not free text

Comparison graphic: Why Checklists Are Better Than Handwriting for AI and Patient Safety. Left panel shows a real handwritten prescription run through AI handwriting recognition, producing low-confidence, error-prone extracted text such as Tab. Ze..dil at 48% confidence, wroid..l at 32%, Cap. Cluaplugin 250 XR at 45%, Syp. Nyr..m at 38%, Tab. Coflen bd. at 52%, NV rinodeal at 41%, and an illegible signature at 60% — with listed risks: missed or wrong medicines, wrong dose or frequency, wrong route, dangerous in high-risk situations, and extra verification burden on staff. Right panel shows the same prescription entered through a structured checklist-based form instead — Patient ID, Date, Diagnosis, and per-medicine dropdown fields for Medicine Name, Form, Dose, Frequency, Duration, and an After Food checkbox — feeding into AI Processes Structured Data (100% readable, standardized, complete, less ambiguity, safe for patient, better for analytics) and then Structured Output usable for EHR, clinical decision support, analytics and research, and audit and quality improvement. A comparison table below contrasts Handwriting+AI (ambiguous input, low confidence, high error risk, extra verification, patient safety at risk) against Checklist+AI (clear and standardized, high accuracy, low error risk, ready for automation, better patient safety), with the key takeaway that AI cannot fix unclear handwriting — the best way to reduce AI transcription errors is to prevent ambiguous data at the source. Closing banner: Better Data In, Better AI Out, Better Decisions, Better Outcomes.
Why checklists beat handwriting for AI and patient safety — structured entry produces clean, reliable data; free-text handwriting produces low-confidence, error-prone extraction

This is the same principle behind every module on this site being built as a structured form or checklist rather than a free-text or handwritten note. AI handwriting recognition can only guess at what a doctor has scrawled — the example above shows extraction confidence as low as 32–52% on a real prescription, with real consequences: missed or wrong medicines, wrong dose or frequency, wrong route. A structured checklist removes the ambiguity at the source. Every field — Patient ID, Diagnosis, Medicine Name, Dose, Frequency, Duration — is captured as a discrete, standardised value the moment it's entered, so there's nothing left for an AI model to misread downstream. That's precisely why the Clinical Examination Templates, the RVC Slip, and the Drug Administration Support Module are all built as checklists and dropdowns rather than free-text boxes: better data in means better AI out, better decisions, and better outcomes.