AI Engineer · ML Engineer · Healthcare AI

Building intelligent
systems for
clinical impact.

End-to-end machine learning pipelines on real-world clinical data — from FDA adverse event prediction to mental health NLP, each deployed as a live interactive application.

5
Live Apps
103K+
Records
88%
Peak Accuracy
APIs Deployed
View Projects Get in Touch
SCROLL DOWN

01

Live Projects

🧪 Clinical Trials · XGBoost · SHAP

Clinical Trial Success Predictor

End-to-end ML on 1,000 real ClinicalTrials.gov records. Predicts whether a trial will produce publishable results. 19 features engineered from 138 raw fields, SHAP explainability identifies enrollment size and trial phase as top predictors.

0.770
AUC-ROC
0.578
F1 Score
81%
Accuracy
1,000
Records
PythonXGBoostSHAPFastAPIStreamlitscikit-learn
Live Demo on HF Spaces
► Resume Line Built end-to-end ML pipeline on 1,000 ClinicalTrials.gov records; 19 features engineered from 138 raw fields; XGBoost (AUC 0.77, F1 0.58); SHAP explainability; FastAPI + Streamlit.
01
📚 NLP · RAG · FAISS · PubMed

PubMed AI Assistant

Biomedical NLP RAG pipeline on 967 PubMed abstracts (2023–2026). FAISS vector store with all-MiniLM-L6-v2 (384-dim) semantic search. Extractive summarization achieving ~92% compression with real PubMed citation links.

967
Papers
3,600+
Chunks
384-dim
Vectors
~92%
Compression
PythonLangChainFAISSSentence TransformersStreamlit
Live Demo on HF Spaces
► Resume Line Built Biomedical NLP RAG pipeline on 967 PubMed abstracts; FAISS + sentence-transformers semantic search; extractive summarization (~92% compression); Streamlit app.
02
💊 FDA FAERS · Drug Safety · SMOTE

FDA Adverse Event Risk Predictor

ML pipeline on 467 real FDA FAERS adverse event reports. XGBoost with SMOTE oversampling for class imbalance. Real-time physician-facing risk scoring — input drug name + patient demographics, get predicted adverse events and risk level.

0.871
AUC-ROC
0.83
F1 Score
467
FAERS Reports
PythonXGBoostSMOTEFastAPIStreamlit
Live Demo on HF Spaces
► Resume Line FDA FAERS ML pipeline; XGBoost + SMOTE on 467 reports (AUC 0.871, F1 0.83); FastAPI + physician-facing Streamlit risk scoring dashboard.
03
🏥 Hospital Readmission · SHAP · Diabetes

Diabetic Patient Readmission Risk

XGBoost model predicting 30-day hospital readmission risk for diabetic patients using 20 engineered clinical features — HbA1c, glucose, BMI, prior admissions, discharge disposition. Per-patient SHAP risk decomposition in dashboard.

0.667
AUC-ROC
0.517
F1 Score
20
Features
PythonXGBoostSHAPPlotlyFastAPIStreamlit
Live Demo on HF Spaces
► Resume Line XGBoost 30-day readmission predictor; 20 clinical features, AUC 0.667; per-patient SHAP decomposition via interactive Streamlit dashboard.
04
🧠 NLP · Mental Health · 7-Class

Mental Health Text Classifier

7-class NLP classifier on 103,000 social media posts: anxiety, bipolar, depression, normal, personality disorder, stress, suicidal. TF-IDF + LR baseline at 88.1% accuracy. DistilBERT fine-tuning script included. Privacy-first with crisis resource integration.

88.1%
Accuracy
88.2%
F1 Score
103K
Posts
7
Classes
Pythonscikit-learnTF-IDFDistilBERTFastAPIStreamlit
Live Demo on HF Spaces
► Resume Line 7-class mental health NLP on 103K posts; TF-IDF + LR (88.1% accuracy, 88.2% F1); DistilBERT fine-tuning; FastAPI + Streamlit, privacy-first + crisis resources.
05

02

Tech Stack

Python
XGBoost
scikit-learn
SHAP
FAISS
LangChain
Sentence Transformers
HuggingFace
FastAPI
Streamlit
Pandas / NumPy
Matplotlib / Seaborn
Plotly
SMOTE
DistilBERT / T5
REST APIs

► Open to Opportunities

Let's build something
that matters.

Seeking AI Engineer, ML Engineer, or Healthcare AI roles.
Available for full-time positions and collaborations.