The bedside reality check.
I still practice medicine. Patients, time pressure, and clinical uncertainty keep my engineering grounded in how care actually happens.
Physician. Engineer.
Building for the bedside.
I build safer clinical AI and FHIR-integrated health software, grounded in patient care.
00 / HOW I WORKEach essential. Each in sync.
From bedside insight to health software.
Code, inspect, iterate.
I build and test the software myself. Data quality, schema boundaries, runtime behavior, and failure modes are engineering decisions with clinical consequences.
From sensor to clinical context.
Recorded ECG and PCG data, BLE experiments, FHIR resources, and clinician-facing interfaces each require careful handoffs. I work across those boundaries without losing the clinical question.
Clinical AI EvalsFHIR R4Active MBBS ClinicianJavaPython + MATLABECG + PCGClinical Informatics
01 / SELECTED WORKFour technical case studies
Clinical systems, engineered from the inside out.
CASE / 01Clinical AI / Safety Evaluation
Vignettes → Inference → Safety intercept → Clinician review
Red-flag omissionCHECK
ContraindicationCHECK
Dosage rangeCHECK
2,500+clinical vignettes
100%synthetic lethal dosage errors intercepted
62%less human review time
Clinical AI Safety & Diagnostic Evaluation
Generic text scores can miss dangerous clinical errors. I built an evaluation path for diagnostic omissions, contraindications, and dosage discrepancies, with clinician review kept in the loop.
REPORTED PROJECT RESULTS2,500+ cases tested; 100% of synthetic lethal dosage errors intercepted; review time reduced 62%.
CASE / 02FHIR R4 / EHR Interoperability
HL7 v2 / ADT + ORUFHIR R4
Patient
Observation
Condition
Medication
Clinical recordNormalize
Clinician workspaceStream
210 → 16scomplex case review
150,000+events per day
0reported drops
FHIR R4 Interoperability & EHR Interface
I developed a path from HL7 v2 feeds to structured FHIR R4 resources, then into a keyboard-driven clinical workspace that makes a patient’s course easier to reconstruct.
For use in India, I consider ABDM’s FHIR R4 profiles, consent-led health-data sharing, and India’s data-protection framework. This project has not undergone formal conformance or legal review.
REPORTED PROJECT RESULTSReview time reduced from 210 to 16 seconds; 150,000+ events/day processed with zero reported drops.
CASE / 03ECG + PCG / Dataset-Based Research
ECG / Electrical activityRecorded data
PCG / Heart soundsRecorded data
Recorded ECG + PCG datasetsPython + MATLAB analysisExploring coronary risk prediction
ECG + PCG Heart-Risk Research
I led an exploration of whether patterns in recorded ECG and heart-sound datasets could help identify coronary artery disease risk. I used Python and MATLAB for the analysis and explored BLE as a possible device connection. The prediction model remains an early research prototype, not a diagnostic tool.
CURRENT STATUSRecorded-data research in progress; clinical validation has not been completed.
CASE / 04Offline-First / Rural Health Systems
FOLLOW-UP COMPLETION84%
Local recordQUEUED
Connection restoredSYNCED
10,000+people tracked
43 → 84%follow-up compliance
72hoffline, zero record loss
Offline-First Public Health Systems
Rural care cannot assume a stable connection. I designed local-first collection and delayed synchronization around field diagnostics and follow-up, with record integrity as the central constraint.
REPORTED PROJECT RESULTS10,000+ people tracked; follow-up compliance rose from 43% to 84%; no record loss during 72-hour outages.
02 / MEDICAL RESEARCHPublished & accepted work
Research beyond
the interface.
Determinants and effects of self-administration with NSAIDs in a tertiary care hospital of Eastern India
A study of why patients self-administer NSAIDs, how they obtain them, and their awareness of adverse effects.
DOI 10.18231/j.ijpp.2020.002Mapping Unmet Surgical Needs in LMICs: A decade of Global Evidence through a scoping review.
Evidence mapping on unmet surgical need across low- and middle-income countries. Accepted by World Journal of Surgery; indexing and citation details are pending.
DR SUMANTA MAJUMDAR / MBBS
Care is personal.
So is the work.
Doctor who codes.
I translate bedside needs into clear product decisions and working software, so clinical intent reaches engineers without getting lost in handoff.
BEYOND THE SCREENSound / Sight
Precision in sound
and sight.
Outside clinical practice and software, I spend my time at the piano and behind a lens. Both sharpen the same instincts I bring to engineering: cadence, restraint, structure, and attention.
03 / FAQDirect answers on how I practice and code.
Questions worth asking.
What changes when a developer also practices medicine?
I can recognize clinical edge cases while writing the system that handles them. Active practice keeps patient safety, workflow friction, and uncertainty close to the implementation.
How does ongoing practice fit with engineering?
Clinical work keeps me grounded in real care. I plan engineering work around those commitments and collaborate asynchronously with clear specifications, code, and decision records.
What is your stance on generative AI in healthcare?
Its outputs need clinical evaluation, deterministic checks where possible, and a human review path. I test missed red flags, contraindications, dosage errors, and whether the interface keeps the clinician in control.
What do you build yourself?
I build clinical AI and FHIR tools, plus React dashboards, web apps, and heatmaps. Java is my strongest language; I also use JavaScript and Python.