Data Scientist / Researcher
Open To Work
Tanurima Halder
The George Institute for Global Health
Durgapur,West Bengal
3
Years of experience
Accomplished Data Scientist at The George Institute for Global Health, adept in developing advanced analytics frameworks and retrieval systems. Expert in Python and model evaluation, with a proven ability to enhance clinical data analysis. Strong collaborator with a focus on achieving impactful results in global health research.
Work History
Data Scientist / Researcher
9 Months- Designed analytics framework for clinical RAG assistant (usage, session duration, per-step latency, error rates logged to SQLite) and 13-metric evaluation framework (faithfulness, hallucination, retrieval recall, PII leakage), achieving 90% response grounding accuracy at 2.6s median latency in 31-query pilot.
- Built retrieval system over unstructured guideline documents (FAISS vector database) and analyzed tradeoff between retrieval precision and guideline alignment across multiple clinical corpora to enhance information retrieval.
Pre-Doctoral Researcher
11 Months- Developed deep learning MRI harmonization models to separate anatomical content from site-specific acquisition artefacts and established privacy-preserving federated-learning frameworks for multi-institution data (ISBI 2025).
- Characterized stage-specific harmonization effects on downstream performance in federated brain age estimation (IEEE ICKG 2025).
- Designed experiments using IIT Bombay laboratory protocols and research methods.
- Collected and organized research data for analysis at IIT Bombay.
- Conducted literature reviews for doctoral research at IIT Bombay.
Junior Research Fellow
10 Months- Co-developed intent-based evaluation metric for legal judgment summaries (intent-F1 tracked human judgment, r = 0.86, single-rater pilot) and hand-annotated dataset of 93 Supreme Court judgments; graph-based summarizer surpassed LetSum and BERT on intent-F1 (0.203 vs. 0.194, 0.187).
Education
M.Tech - Artificial Intelligence and Machine Learning
GPA: 7.59/10
B.Tech - Computer Science and Engineering
GPA: 8.69/10
Skills
Data Science & ML: Python
SQL
Pandas
NumPy
Scikit-learn
PyTorch
TensorFlow
Model evaluation
Weak supervision
Federated learning
NLP & GenAI: HuggingFace Transformers
NLP
LLMs
RAG
LangChain
FAISS / vector search
prompt engineering
Domains & Tools: Clinical support
medical imaging imagingext; DocLegal analysisLinux
SQLite
AWS
Azure
Databricks
Machine learning
Retrieval systems
Federated learning
Model evaluation metrics
Personal Information
Title: Data Scientist | NLP, Healthcare Analytics & Model Evaluation
SELECTED PROJECTS
- Intent-Aware Legal Judgment Summarization (2026; Python, PyTorch, Scikit-learn): weak supervision and a multi-task BiGRU tagger on the public IN-Abs corpus (7,127 judgments). At equal length, the intent-conditioned summarizer kept 81% of weakly labelled central sentences vs. 66% for TextRank and improved ROUGE-2 by 4.5 points. [GitHub]
- LLM Reliability Study (under review, ACL Rolling Review, Oct 2026): observational study of 68,476 trials (six instruction-tuned models, four QA domains) plus a controlled follow-up. Errors track loss of the answer rather than loss of context; related but insufficient context drives fabrication.
PUBLICATIONS
- Halder et al. When to Harmonize? Evaluating Stage-Specific Harmonization in Federated Brain Age Estimation. IEEE ICKG 2025.
- Galada, Halder et al. PRISM: Privacy-Preserving Inter-Site MRI Harmonization via Disentangled Representation Learning. ISBI 2025.
- Nandy, Halder et al. Lexical Complexity Prediction using Transformer-based Language Models. ACL SemEval 2021.
Timeline
Data Scientist / Researcher
The George Institute for Global Health
07.2025 - 04.2026Read More
Pre-Doctoral Researcher
IIT Bombay
07.2024 - 06.2025Read More
Junior Research Fellow
IIT Kharagpur
09.2020 - 07.2021Read More
IIT Hyderabad
M.Tech from Artificial Intelligence and Machine Learning
Read More
Jalpaiguri Government Engineering College
B.Tech from Computer Science and Engineering
Read More