Narendra Kalam — Data Scientist / ML Engineer

Selected Work

Systems, not scripts.

Every model below went through the same governance pipeline before deployment — the same standard a production ML team would hold itself to.

✓Metric threshold gate

→

✓Generalization gap check

→

✓Explainability & drift monitoring

Credit Card Fraud Detection

BFSI · Fintech

ExtraTrees

0.896F1 Score

284KTransactions

0.967ROC-AUC

Real-time fraud scoring on heavily imbalanced data, with cost-sensitive evaluation and live monitoring.

ExtraTreesFastAPIStreamlit

GitHub ↗ Live demo ↗

Credit Risk Decision Engine

BFSI · Lending

LightGBM

0.974F1 Score

0.999ROC-AUC

0.982KS Stat

A 3-tier APPROVE / REVIEW / DECLINE engine with isotonic calibration and per-prediction SHAP explainability.

LightGBMFastAPISHAPDocker

GitHub ↗ Live demo ↗

Telco Customer Churn

Telecom · BFSI

CatBoost

0.634F1 Score

0.839ROC-AUC

0.731Recall

A RETAIN / OUTREACH / URGENT intervention engine with revenue-impact-weighted decisioning.

CatBoostFastAPIMLflow

GitHub ↗ Live demo ↗

House Price Valuation

Real Estate

CatBoost

0.905R² Score

$20.1KRMSE

8.9%MAPE

A 4-band automated valuation model — AFFORDABLE / MID / PREMIUM / LUXURY — mirroring bank LTV assessment.

CatBoostSHAPFastAPI

GitHub ↗ Live demo ↗

Store Sales Forecasting

Retail · Time Series

LightGBM Ensemble

0.976R² Score

0.374RMSLE

12.1%WAPE

Forecasts for 54 stores × 33 product families, with walk-forward validation and zero data leakage.

LightGBMProphetSHAP

GitHub ↗ Live demo ↗

Grid Energy Demand Forecast

Energy · Utilities

ElasticNet

0.976R² Score

712 MWRMSE

96.2%Band Acc.

A 4-tier dispatch engine for grid operators — from normal operation to emergency procurement.

ElasticNetPSI DriftStreamlit

GitHub ↗ Live demo ↗

Stock Price & Risk Forecasting

Quant Finance

Ridge

0.80Sharpe Ratio

53.4%Dir. Accuracy

63Features

A VaR-gated BUY/HOLD/SELL signal engine — hard risk rules always override the model, never the reverse.

RidgeLSTM/GRURisk Engine

GitHub ↗ Live demo ↗

Resume Screener AI

HR Tech · NLP

LightGBM

0.761F1 Macro

94.2%Top-3 Acc.

0.979ROC-AUC

A TF-IDF match engine ranking candidates against job descriptions — SHORTLIST / REVIEW / REJECT, ATS-style.

NLPTF-IDFFastAPI

GitHub ↗ Live demo ↗

Aspect-Based Sentiment Analysis

E-Commerce / Banking · NLP

RidgeClassifier

0.621Macro-F1

0.823ROC-AUC

73.6%Accuracy

Aspect-level sentiment classification for laptop and restaurant domains, identifying sentiment per product feature rather than overall tone.

NLPScikit-LearnFastAPI

GitHub ↗ Live demo ↗

Fake News Detector

Media Tech / Gov Tech · NLP

XGBoost

0.999F1 Score

1.000ROC-AUC

44KArticles

Credibility classification system with a full feature engineering pipeline (TF-IDF + text statistics) and PSI drift monitoring.

XGBoostMLflowFastAPI

GitHub ↗ Live demo ↗

BC5CDR Clinical NER

Biomedical NLP

BioBERT

0.885F1 Score

0.924Chemical F1

15K+Sentences

Named entity recognition for chemical and disease mentions in clinical text, fine-tuned on the BC5CDR benchmark corpus.

BioBERTNERFastAPI

GitHub ↗ Live demo ↗

News Topic Modeling

Media Analytics · NLP

LDA (Gensim)

0.6225Coherence

0.92Diversity

200KArticles

Unsupervised topic discovery across 40+ news categories, surfacing latent themes without manual labeling.

LDAGensimStreamlit

GitHub ↗ Live demo ↗

Chest X-Ray Diagnosis

Healthcare AI · Medical Imaging

DenseNet121

0.786Mean AUC

14Pathologies

112KX-rays

A CNN-based radiology triage system with clinical severity tiers and Grad-CAM explainability — reduces routine review load by 30–40%.

PyTorchGrad-CAMFastAPI

GitHub ↗ Live demo ↗

Data Scientist building production-grade ML systems.

Systems, not scripts.

Currently building.

What's under the hood.

Modeling

MLOps & Deployment

Domains

Education & Certifications