M.Tech in Artificial Intelligence | Indian Institute of Technology Jodhpur
Research Focus: Parameter-Efficient Fine-Tuning, Large Language Models, Multi-Modal Representation Learning
I am a graduate researcher at IIT Jodhpur specializing in efficient adaptation methods for large language models and multi-modal representation learning. My work sits at the intersection of natural language understanding, parameter-efficient fine-tuning (PEFT), and scalable ML systems—with a particular emphasis on persona alignment, quantization-aware training, and contrastive learning frameworks.
Prior to my M.Tech, I completed my M.Sc. in Mathematics, which provides a rigorous foundation in optimization theory, linear algebra, and statistical inference that informs my approach to deep learning research.
Current Interests:
- Low-rank adaptation (LoRA/QLoRA) for LLM alignment and behavioral steering
- Multi-modal fusion architectures for large-scale clustering
- Efficient inference pipelines and serverless GPU deployment
- Evaluation frameworks for generative models (perplexity, similarity metrics)
Fine-tuning Qwen-2.5 for rigorous persona adoption using QLoRA (4-bit NF4 quantization)
- Implemented low-rank adaptation (r=16) on Qwen-2.5-7B with BitsAndBytes quantization
- Designed dual-metric evaluation: Conditional Perplexity & Jaccard Similarity for persona steering
- Achieved perplexity reduction from 45.31 → 7.52; deployed via Modal serverless API
PythonPyTorchHugging Face (PEFT, TRL)QLoRABitsAndBytesModalFastAPI
End-to-end deep learning pipeline fusing demographic, textual (S-BERT), and behavioral (LSTM) modalities
- Engineered multi-modal encoder with dual-loss training (Contrastive + Clustering)
- Processed 1.36M+ customer embeddings with GPU-accelerated HDBSCAN
- Achieved 0.901 Silhouette Score on large-scale segmentation benchmark
PythonPyTorchSentenceTransformersLSTMcuML (RAPIDS)HDBSCAN
Languages
Deep Learning & ML Frameworks
LLM & GenAI
Infrastructure & Tools
| Degree | Institution | Performance | Year |
|---|---|---|---|
| M.Tech (Artificial Intelligence) | IIT Jodhpur | 9.5 CGPA (Sem-1) | 2025–2027 |
| M.Sc. (Mathematics) | University of North Bengal | 6.46 CGPA | 2021–2023 |
| B.Sc. (Mathematics) | Vidyasagar University | 8.62 CGPA | 2018–2021 |
GATE Data Science & AI 2025: AIR 226 (Score: 717)
Teaching Assistant — IIT Jodhpur
Introduction to Machine Learning (July 2025 – December 2025)
For research collaborations, discussions on LLM alignment, or PEFT methodologies:
- Email:m25csa023@iitj.ac.in | saupriyadip571@gmail.com
- Portfolio:priyadipsau.in
"The goal is not to build models that perform well, but models that generalize wisely."