AAKASH@portfolio:~$ ~/home
$ HOME

AAKASH_OS v2.6 ready.

System loaded. Navigate sections via sidebar or keys [1-8].

Last login: on tty1


##### ##### ## ## ## ## ####### ####### ## ## ## ## ## ## ## ##
AAKASH ANNADURAI
AI-ML ENGINEER

AI/ML Engineer candidate specializing in Machine Learning, anomaly detection, and deep learning. Pursuing PGP in AI at VIT Bangalore. Open to internships and junior roles.


[JAN 2025 – MAY 2025] Junior SDE · Three Dots Software Development
└─React frontend for warehouse management system (3+ workflows)
└─Delivered 4 Agile sprints with team of 5 · Git · modular design patterns

[2025–PRESENT] PGP Artificial Intelligence · VIT Bangalore · CGPA: 10.0
[2019–2025] B.Tech Chemical Engineering · SASTRA University · CGPA: 6.3

Unsupervised Network Intrusion Detection System 2026
STACK: [Python] [PyTorch] [VAE] [Isolation Forest] [HDBSCAN] [SHAP] [Flask] [NSL-KDD] [UNSW-NB15] [ADWIN] [MITRE ATT&CK]
  • Stacked ensemble (VAE + Isolation Forest [w=0.835] + HDBSCAN): ROC-AUC 0.716, Precision 64.3%, Specificity 82.6% on 125,973-record UNSW-NB15
  • Zero attack labels used — pure unsupervised detection
  • Deployed Flask app with SHAP explainability, ADWIN drift monitoring, and MITRE ATT&CK alert mapping
F1 Pitstop Strategy Simulator 2025
STACK: [Python] [FastF1] [Polynomial Regression] [Random Forest] [Gradient Boosting] [Streamlit]
  • Physics Correction Engine isolated tyre degradation from telemetry noise
  • Polynomial Regression R²=0.72 / RMSE=0.28 — outperformed RF (0.68) and GBM (0.65)
  • 3 seasons of F1 data · interactive Streamlit dashboard · 1,000+ pit-stop scenarios
Loan Repayment Prediction 2025
STACK: [Python] [Random Forest] [Logistic Regression] [Stacking Ensemble] [Scikit-learn]
  • Random Forest 89.6% accuracy on 20,000-record dataset vs Logistic Regression (88.0%)
  • Full pipeline: OneHotEncoding · StandardScaler · 80/20 train-test split
Unsupervised Voice Clustering 2026 — Publication
STACK: [Python] [WaveLM] [Wav2Vec 2.0] [OpenSMILE] [UMAP] [HDBSCAN]
  • Silhouette Score 0.627 · DB-Index 0.358
  • UMAP→HDBSCAN pipeline for speaker embedding clustering
  • Accepted — Conference Presentation Pending, April 2026
Lifestyle Segmentation 2025
STACK: [Python] [K-Means] [PCA] [Silhouette Analysis] [HDBSCAN] [EDA]
  • K-Means + PCA segmented 4 lifestyle profiles from obesity behavioral data
  • Silhouette Score 0.652 · 2 PCA components explaining 67% variance
Sleep Efficiency Prediction 2024
STACK: [Python] [Minitab] [Multivariate Linear Regression] [ANOVA] [VIF]
  • Multivariate regression R²=80.29% · 7 predictors via stepwise selection · all p<0.05
  • Full diagnostics: VIF<5, Lack-of-Fit p=0.099 · Awakenings confirmed top predictor
$ ls ~/github/repos

fetching repositories from github.com/Aakash-Annadurai...

[2026] "Unsupervised Voice Clustering Using WaveLM, Wav2Vec 2.0, and OpenSMILE with UMAP and HDBSCAN"
Vijayakumar · Kauluri · Aakash Annadurai · Gowri G.S. — VIT Bangalore
METRICS: WaveLM Silhouette 0.627 · DB-Index 0.358

MACHINE LEARNING
Supervised & Unsupervised Learning · Classification · Regression
Clustering (K-Means, HDBSCAN) · Anomaly Detection · Ensemble Methods
Dimensionality Reduction (PCA, UMAP) · Explainable AI (SHAP)
DEEP LEARNING
VAE · PyTorch · Neural Networks · Latent Space Modeling
Reconstruction-based Anomaly Detection
METRICS & EVALUATION
ROC-AUC · F1-Score · Precision/Recall · R² · RMSE · MAE
Silhouette Score · Davies-Bouldin Index
LIBRARIES & TOOLS
Scikit-learn · PyTorch · Pandas · NumPy · Streamlit · Flask
React · FastF1 · SHAP · HDBSCAN · Statsmodels · Git · Docker
LANGUAGES & DATABASES
Python · JavaScript · Java · MongoDB · SQLite · SQL
CONCEPTS
Zero-Day Detection · Concept Drift (ADWIN) · MITRE ATT&CK
Statistical Inference · ANOVA · Agile/Scrum

$ cat contact.info
$ cat links.info
$ cat phone.info
[PHONE] +91 86105 23216

EOF — end of contact output

[~] home [/about] profile [/exp] experience [/edu] education [/projects] projects [/pub] publications [/skills] skills [/contact] contact
SCROLL: 0% home [↑↓] navigate