I'm a final-year IT student at NIT Jalandhar (graduating 2026), and I spend most of my time building things at the crossroads of AI/ML engineering, NLP research, time-series forecasting, and open-source development.
I care about practical ML that actually solves real problems: cutting renewable energy forecast error by nearly half, detecting hate speech fairly across demographics, and shipping open-source tools people use on their Linux desktops.
veeraj = {
"role" : "AI/ML Engineer | NLP Researcher",
"school" : "NIT Jalandhar, B.Tech IT (2026)",
"intern" : "Greenko Group, Wind Power Forecasting",
"research" : "COUNTER-HATE, Multimodal Hate Detection",
"os" : "Arch Linux 🐧",
"seeking" : "AI/ML | NLP | ML Research Engineer roles",
}🎯 Currently working on:
- 🔭 Time-series forecasting pipelines for renewable energy scheduling
- 🧪 Multimodal NLP research and fairness-aware AI systems
- 🛠️ CryptoStats widget for KDE Plasma 6
- 📖 RAG systems, vLLM, and production LLM deployment
💬 Ask me about: NLP LLMs Time-Series PyTorch Arch Linux
 |
 |
 |
 |
| Wind power scheduling for 30+ energy plants |
Ridge + LightGBM ensemble 15-min interval forecasting |
Multimodal hate speech bias-aware detection |
KDE Plasma 6 widget QML + curl |
| Domain | Tools & Experience |
|---|---|
| 🤖 ML & Deep Learning | PyTorch · TensorFlow · scikit-learn · LightGBM · XGBoost |
| 🗣️ NLP & Multimodal AI | HateBERT · CLIP · Transformers · RAG · vLLM · Ollama |
| 📈 Time-Series | 15-min interval · Renewable energy · Ensemble methods |
| 🐧 Open-Source / Linux | KDE Plasma · Arch Linux · QML · Bash |
| 🦠 Research | Fairness eval · Counterfactual augmentation · EEG/BCI |
|
Multimodal Counterfactual Augmentation for Bias-Aware Hate Detection Built a multimodal hate speech detection system using LLM-generated counterfactual rewrites and synthetic images. Combines adversarial training, multimodal fusion, and fairness evaluation to reduce demographic bias in model predictions.
|
Renewable Energy Time-Series Forecasting Forecasted wind output for 30+ plants at 15-minute resolution. A Ridge + LightGBM ensemble cut MAE from 7.28 to 3.70 MWh. Shipped with a Streamlit operator dashboard and an Ollama RAG chatbot for plant teams.
|
|
KDE Plasma 6 Cryptocurrency Widget Real-time crypto pricing right on the Linux desktop. Built with QML and curl, keeping overhead minimal with clean KDE Plasma 6 integration.
|
EEG Motor Imagery Classification (Research / WIP) Experimental CNN+LSTM for EEG-based motor imagery BCI. Currently being refactored to fix data leakage in preprocessing. Not a finalized model.
|
|
Semantic Shift Detection in Text and Prompts Detects semantic shifts, claim changes, and tone drift across document and prompt versions. Useful for LLM prompt tracking and policy analysis.
|
Medical Imaging with Deep Learning CNN-based brain tumor classification (Glioma, Meningioma, Pituitary) from MRI scans.
|
Languages
AI / ML
NLP / LLMs
Data / Visualization
Tools & Platforms
| 🥇 Achievement | Details |
|---|---|
| 🏫 AI/ML Engineer Intern | Greenko Group, Renewable Energy AI |
| ⚡ Forecasting Impact | MAE reduced from 7.28 to 3.70 MWh |
| 🔬 NLP Research | COUNTER-HATE, Multimodal Bias-Aware Detection |
| 🐧 Open Source | CryptoStats, KDE Plasma 6 widget with 2+ stars |
| 🎓 Education | B.Tech IT, NIT Jalandhar (2026) |
| 🏋️ Kaggle | Active ML practitioner, kaggle.com/veeraj16 |
Building practical AI systems, research-driven ML workflows, and open-source tools.