I am an AI/ML engineer and researcher specializing in large language models, generative AI, and data-driven optimization. I have hands-on experience building RAG systems, fine-tuning transformer models, designing scalable ML pipelines, and analyzing model performance under real-world constraints. With a strong foundation in operations research and machine learning, I focus on creating robust, efficient, and interpretable AI solutions for complex problems. I primarily work with Python, PyTorch, Hugging Face, and modern AI tooling to deliver production-ready results.…

Viraj Vardhan

I am an AI/ML engineer and researcher specializing in large language models, generative AI, and data-driven optimization. I have hands-on experience building RAG systems, fine-tuning transformer models, designing scalable ML pipelines, and analyzing model performance under real-world constraints. With a strong foundation in operations research and machine learning, I focus on creating robust, efficient, and interpretable AI solutions for complex problems. I primarily work with Python, PyTorch, Hugging Face, and modern AI tooling to deliver production-ready results.…

Available to hire

I am an AI/ML engineer and researcher specializing in large language models, generative AI, and data-driven optimization. I have hands-on experience building RAG systems, fine-tuning transformer models, designing scalable ML pipelines, and analyzing model performance under real-world constraints. With a strong foundation in operations research and machine learning, I focus on creating robust, efficient, and interpretable AI solutions for complex problems. I primarily work with Python, PyTorch, Hugging Face, and modern AI tooling to deliver production-ready results.

See more

Skills

Experience Level

Prompt Engineering
Expert
Python
Expert
PyTorch
Expert
R
Expert
Keras
Expert
TensorFlow
Expert

Language

English
Fluent

Work Experience

Graduate Researcher at Centre for Reliable Wealth Transition, University of Ottawa, Canada
July 1, 2025 - Present
Conducted extensive empirical analysis to characterize neural scaling behavior across multiple LLMs, examining performance trends under varying model sizes, dataset scales, and random initialization seeds. Evaluated robustness of scaling models across benchmark datasets and experimental configurations, identifying consistent patterns in data efficiency and generalization trade-offs for fine-tuning. Formulated a stochastic optimization framework to determine the optimal allocation of training and validation data under a fixed data-budget, maximizing performance while accounting for statistical uncertainty. Derived and analyzed a confidence-adjusted objective function incorporating standard error terms and critical values, to balance performance estimation bias and variance during data allocation.
Research Assistant at Industrial Engineering and Operations Research Lab, IIT Delhi, India
January 1, 2025 - May 1, 2025
Designed an AI-driven image processing pipeline to automate 3D bone reconstruction from high-resolution CT scans, integrating voxel interpolation, segmentation, and volume rendering for precise anatomical modeling. Implemented deep learning-based fracture detection using a convolutional architecture trained on ~100,000 CT slices, optimizing performance through class rebalancing, augmentation, and regularization. Developed a heuristic shape matching algorithm using point-set distance minimization to detect cortical discontinuities by comparing 3D reconstructions of healthy and fractured bone surfaces. Built an interactive 2D–3D visualization framework, integrating voxel-based rendering and real-time slice mapping to facilitate spatial correlation analysis between CT data and reconstructed anatomical structures.
Research Assistant at Industrial Engineering and Operations Research Lab, IIT Delhi, India
August 1, 2024 - December 1, 2024
Implemented advanced image processing techniques, including Local Binary Patterns (LBP), Gabor filters, and color space conversion, to enhance texture and color extraction, optimizing segmentation for Phenocam images. Developed and optimized clustering algorithms, including Kernel K-means, Gaussian Mixture Model, Normalized Cut, and computationally efficient methods like SLIC and Felzenszwalb's segmentation. Developed a space-optimized approach for Large Scale Kernel K-means, reducing memory usage by on-the-fly computations, enabling the algorithm to handle massive datasets while maintaining clustering accuracy. Developed a neural network-based pairwise clustering model using pixel-wise RGB and spatial features, optimizing similarity with KL divergence and hinge loss to enhance region-based clustering precision.
Research Assistant at Lucy Family Institute, University of Notre Dame, USA
June 1, 2024 - July 1, 2024
Developed a RAG framework integrating Sentence-Transformer all-MiniLM-L6-v2 embeddings with FAISS-based dense vector indexing to enable semantic retrieval and context-aware text generation across multi-domain. Fine-tuned a DistilBART-CNN-12-6 transformer within an institutional pipeline, optimizing coherence and factual grounding through retrieval fusion and cross-attention relevance scoring. Designed a scalable data-engineering workflow to synthesize and preprocess unstructured text repositories, employing tokenization, metadata alignment, and embedding normalization for high retrieval precision. Deployed end-to-end system on the Lucy Family Institute’s compute infrastructure, supporting secure, interactive querying and evaluation via a Gradio-based web interface.
Research Intern at Global Internship Program, University of Auckland, New Zealand
July 1, 2023 - August 1, 2023
Built a modular GPT-style autoregressive language modeling framework from first principles, enabling controlled experimentation across tokenization strategies, attention variants, and training dynamics. Implemented full Transformer blocks (residual pathways, LayerNorm, GELU) with detailed instrumentation to analyze optimization stability, gradient behavior, and representation quality in deep autoregressive models. Fine-tuned pretrained language models for both classification and instruction-following tasks, evaluating supervision styles, prompt formatting, and instruction conditioning on downstream performance. Designed end-to-end automated conversational evaluation pipelines to benchmark model quality and consistency, analyzing decoding strategies and sampling variability across inference configurations.
Research Assistant at Data Analytics & Network Science Lab (DANeS), IIT Patna, India
May 1, 2023 - July 1, 2023
Developed a Spatio-Temporal Graph Convolutional Network (STGCN) model to analyze and predict traffic patterns, effectively capturing temporal dependencies and spatial correlations from extensive traffic datasets. Optimized data preprocessing pipelines by implementing feature scaling and segmentation techniques, transforming raw traffic data into structured input suitable for model training and evaluation. Utilized Chebyshev polynomial approximation and scaled Laplacian matrices to enhance graph convolution operations, improving model's ability to capture complex spatial relationships in traffic data. Implemented a knowledge distillation strategy to refine model performance, enabling a compact student model to assimilate knowledge from a larger teacher model, achieving high prediction accuracy with reduced computation.
Research Assistant at Centre for Transportation Systems (CTRANS), IIT Roorkee, India
December 1, 2022 - January 1, 2023
Performed exploratory data analysis (EDA) on a large bus network dataset (6,000 vehicles & 8,000 routes) using Python libraries to identify spatiotemporal inefficiencies in public bus operations. Engineered automated spatiotemporal data pipelines integrating statistical and probabilistic modeling to detect systematic demand–supply imbalances and temporal inefficiencies in transit operations. Designed an automated route management system using heuristic-based algorithms for dynamic bus-to-route allocation, integrating real-time demand forecasting and operational constraints to optimize fleet efficiency. Developed a modular experimentation framework to iteratively evaluate routing policies under varying demand scenarios.
Graduate Researcher at Centre for Reliable Wealth Transition, University of Ottawa
July 1, 2025 - Present
Robust Data Collection and Allocation for Fine-Tuning Large Language Models. Conducted empirical analyses of neural scaling across LLMs, evaluating performance trends under varying model sizes, dataset scales, and random seeds. Assessed robustness across benchmark datasets and configurations to understand data efficiency and generalization trade-offs in fine-tuning. Formulated a stochastic optimization framework to allocate training and validation data within a fixed budget, maximizing performance under uncertainty. Derived a confidence-adjusted objective incorporating standard errors and critical values to balance bias and variance in data allocation.
Research Assistant at Industrial Engineering and Operations Research Lab, IIT Delhi
January 1, 2025 - May 1, 2025
AI-Based Medical Image Reconstruction and Structural Damage Detection. Designed an AI-driven pipeline to automate 3D bone reconstruction from high-resolution CT scans, integrating voxel interpolation, segmentation, and volume rendering. Implemented a CNN-based fracture detection model trained on ~100,000 CT slices with class balancing, augmentation, and regularization. Developed a heuristic shape-matching algorithm using point-set distance minimization to detect cortical discontinuities, and built an interactive 2D–3D visualization framework for spatial correlation analysis between CT data and reconstructed anatomy.
Research Assistant at Industrial Engineering and Operations Research Lab, IIT Delhi
August 1, 2024 - December 1, 2024
Intelligent Image Processing for Phenological Monitoring. Implemented Local Binary Patterns and Gabor filters to enhance texture and color features for segmentation of Phenocam images. Explored clustering algorithms including Kernel K-means, Gaussian Mixture Models, Normalized Cuts, and efficient methods like SLIC and Felzenszwalb. Developed a space-optimized Large-Scale Kernel K-means and a neural network-based pairwise clustering model using pixel-wise RGB and spatial features to improve region-level clustering.
Research Assistant at Lucy Family Institute, University of Notre Dame
June 1, 2024 - July 1, 2024
RAG Framework for Contextual Knowledge Synthesis. Built a Retrieval-Augmented Generation framework integrating Sentence-Transformer all-MiniLM-L6-v2 embeddings with FAISS-based dense indexing to enable semantic retrieval and context-aware text generation across domains. Fine-tuned a DistilBART-CNN-12-6 transformer within institutional pipelines to improve generative coherence and grounding. Designed scalable data-engineering workflows to preprocess unstructured text, including tokenization, metadata alignment, and embedding normalization, and deployed end-to-end on secure compute infrastructure with a Gradio web interface for querying and evaluation.
Research Intern at Global Internship Program, University of Auckland
July 1, 2023 - August 1, 2023
Large Language Model Research & Evaluation Framework. Built a modular GPT-style autoregressive framework to experiment with tokenization, attention variants, and training dynamics. Implemented full Transformer blocks with instrumentation to analyze optimization stability and representation quality. Fine-tuned pretrained LMs for classification and instruction-following tasks, and designed automated conversational evaluation pipelines to benchmark model quality and consistency across inference configurations.
Research Assistant at Data Analytics & Network Science Lab (DANeS), IIT Patna
May 1, 2023 - July 1, 2023
Deep Neural Network for Urban Traffic Flow Forecasting. Developed a Spatio-Temporal Graph Convolutional Network (STGCN) to analyze and predict traffic patterns, capturing temporal and spatial dependencies. Optimized data preprocessing, applied Chebyshev polynomial approximations and scaled Laplacian matrices to enhance graph convolutions, and implemented a knowledge distillation strategy to create a compact student model with high accuracy.
Research Assistant at Centre for Transportation Systems (CTRANS), IIT Roorkee
December 1, 2022 - January 31, 2023
Optimization of Public Transit Efficiency Through Dynamic Bus Allocation. Conducted EDA on a large bus network (6,000 vehicles, 8,000 routes) to identify spatiotemporal inefficiencies. Engineered automated spatiotemporal data pipelines combining statistical and probabilistic models to detect demand-supply imbalances and timing inefficiencies. Designed an automated route management system using heuristic algorithms with real-time demand forecasting and constraints to optimize fleet utilization.

Education

M.S. in Systems Engineering at University of Ottawa
September 1, 2025 - January 4, 2026
B.Tech in Production and Industrial Engineering at Indian Institute of Technology (IIT) Delhi
November 1, 2021 - May 1, 2025

Qualifications

Add your qualifications or awards here.

Industry Experience

Education, Software & Internet, Professional Services, Media & Entertainment, Life Sciences
    paper Large Language Model Research & Evaluation Framework

    • Built a modular GPT-style autoregressive language modeling framework from first principles, enabling
    controlled experimentation across tokenization strategies, attention variants, and training dynamics.
    • Implemented full Transformer blocks (residual pathways, LayerNorm, GELU) with detailed instrumentation to analyze optimization stability, gradient behavior, and representation quality in deep autoregressive models.
    • Fine-tuned pretrained language models for both classification and instruction-following tasks, evaluating
    supervision styles, prompt formatting, and instruction conditioning on downstream performance.
    • Designed end-to-end automated conversational evaluation pipelines to benchmark model quality and
    consistency, analyzing decoding strategies and sampling variability across inference configurations.