Clinical AI evaluation, calibration theory, IoT in healthcare, and LLM behaviour — papers I have authored or contributed to.
- 09
- Papers
- 06
- Published | Presented
- 03
- In review
Severity-Weighted Calibration Error for Reliability Assessment Under Outcome-Severity Imbalance
IEEE Access
2026Aggregate calibration scores mask failures in high-acuity subgroups. Per-band isotonic recalibration consistently fixed this; global recalibration did not.
Aman Chandra H, K.C. Narendra
Intravenous Bag Monitoring and Alert System — A Prototype
Springer Nature · CCIS
2025Real-time IV fluid monitoring using microcontroller-based non-contact capacitive sensors, integrated with an IoT platform for centralised alerting. Presented at ICBEST 2025.
Publisheddoi:10.1007/978-3-032-14908-4_45Aman Chandra H et al.
Severity-Aware Evaluation of ICU Mortality Prediction Using Risk-Weighted Error Metrics (RWEM)
BMC Medical Informatics · Springer Nature
2025Introduces RWEM — a metric that penalises missed deaths by patient severity. Evaluated across 5 models and 3 critical care databases.
Under Peer Reviewdoi:10.21203/rs.3.rs-9429272/v1- Presented · 18–22 Aug 2026
Prevalence Shift and ICU Mortality Miscalibration Across Institutions
ICETCI 2026 · Mahindra University, Hyderabad
2026Zero-label Bayesian correction for prevalence shift achieving 56–65% calibration error reduction across 185 hospitals without model retraining.
Aman Chandra H, Stuti Shivhare, Suyash Sahu
- Presented · 18–22 Aug 2026
Language as a Hidden Variable: Measuring Behavioral Divergence in Multilingual Large Language Models
ICETCI 2026 · Mahindra University, Hyderabad
2026A controlled multilingual study introducing BDS and showing cross-language behavioral divergence up to 2.5x within-language variation.
Presenteddoi:10.21203/rs.3.rs-10218350/v1Aman Chandra H, Sameer Kulkarni, S Nithin, Khushboo Pandey
- Presented · 18–22 Aug 2026
NOMOS: RAG Framework for Jurisdiction-Specific Indian Legal QA
ICETCI 2026 · Mahindra University, Hyderabad
2025Privacy-preserving legal QA over 2.32M characters of Indian statutory text using local LLM inference, ChromaDB, and React + Firebase. Outperformed GPT-4 on citation accuracy.
PresentedSameer Kulkarni, Aman Chandra H et al.
- Latest · Presented · 18 Sep 2026
Clinical Risk-Weighted Evaluation of ICU Mortality Prediction Models (CRWS)
ICAIHC 2026 · NIT Raipur
2025A cost-sensitive evaluation framework for ICU mortality, penalising missed deaths over false alarms. Evaluated across 5 model classes on 3 independent critical care databases.
PresentedAman Chandra H, Abhijna Shivaprakash et al.
Does Model Compression Amplify Clinical Bias? A Subgroup Fairness Audit of Quantized ICU Risk Models on MIMIC-IV
IEEE INDICON 2025
2026Audits whether INT8 quantization, pruning, and knowledge distillation amplify subgroup fairness gaps in ICU AKI prediction — and uncovers a quantization failure mode caused by outlier EHR values.
SubmittedAman Chandra H
Graph-Based Polypharmacy Risk Networks for Adverse Event Prediction in ICU Patients: A MIMIC-IV Analysis
IEEE DELCON, BITS Pilani
2026Compares graph embeddings against tabular features for predicting AKI, delirium, and bleeding from ICU polypharmacy patterns — and catches a label-leakage bug that was inflating results before correction.
SubmittedAman Chandra H