Sarthak.
I transform ideas into AI-driven realities.
A passionate Machine Learning Engineer. I specialize in cutting-edge AI technologies like LLMs and Stable Diffusion, creating innovative solutions that redefine possibilities.
Hi, my name is
About Me
I am a dedicated and driven individual who recently graduated with a Masters of Data Science degree from University of Adelaide, specializing in cutting-edge GenAI technologies such as GANs, Stable Diffusion and LLMs. With a strong academic background and practical experience in developing tools for early diagnosis, as well as mentoring fellow students, I am passionate about leveraging AI to solve real-world challenges and drive innovation. I am open to work.
I am currently working on a project that will be used to build a personal assistant but which combines Stable Diffusion along with LLMs.
Here are a few technologies I've been working with recently:- Diffusion Models
- Generative Adversarial Networks
- Large Language Models
- HuggingFace
- Docker
- PyTorch
Experience
- Researched and optimized diffusion models for high-quality image and video synthesis.
- Accelerated ML pipelines, reducing generation time via model distillation, quantization, and caching.
- Researched on enhancing text-to-image and video generation using latent space optimization and diffusion techniques.
- Fine-tuned deep learning models for efficient deployment in VFX production.
- Collaborated with cross-functional teams to integrate AI-driven tools into existing workflows.
- Conducted comprehensive literature reviews to remain updated on the latest advancements in LLMs, automated program repair, and associated optimization techniques.
- Fine-tuned Large Language Models (LLMs) such as Codellama, PLBART, Refact, CodeGen, and CodeT5 for Automated Program Repair (APR) tasks, employing techniques including PEFT (Parameter Efficient Fine-Tuning), LoRA (Low-Rank Adaptation), and int-8 optimization.
- Implemented and optimized LLMs using state-of-the-art methodologies, ensuring compatibility with resource-constrained environments through various optimization methods like Gradient Checkpointing, int-8 optimization, etc.
- Utilized SLURM to manage and distribute computing tasks across multiple GPUs, optimizing model training and experimentation efficiency.
- Played a key role in securing additional funding for the project, leveraging my expertise and contributions to demonstrate its potential and value to stakeholders.
- Collaborated closely under the guidance of supervisors to develop expertise in Real-Time Neural Style Transfer on Videos.
- Successfully navigated significant challenges, including conducting extensive literature reviews, implementing state-of-the-art methods, and enhancing them to ensure temporal consistency across consecutive frames in live camera feed videos during style transfer.
- Acquired comprehensive knowledge of research areas such as Deep Learning Optimization Techniques, including Knowledge Distillation and Quantization. Utilized libraries such as TensorRT and Pytorch-JIT to accelerate inference time while performing style transfer.
- Investigated the impact of Automatic Mixed Precision on the training and inference speeds and GPU utilization of PyTorch models for different class of GPUs
- Collaborated closely within an 8-member team to develop a Customer Relationship Management (CRM) system for a logistics enterprise.
- Engaged in a variety of daily responsibilities, such as designing and developing the database, managing the migration of changes, creating and testing REST APIs, and facilitating the handover of these components to the frontend team.
- Additionally, contributed to user clustering efforts and the prediction of purchase order patterns among users, leveraging existing data provided by the logistics company.
- Played a role in designing a forecasting algorithm to enhance understanding of seasonal logistics bookings by analyzing historical data furnished by the company.
- Engaged in daily activities that encompassed researching established Computer Vision architectures and exploring avenues for their enhancement.
- Conducted research focused on image classification and segmentation tasks within the medical domain, specifically addressing areas such as Brain Tumor Segmentation and Glaucoma Classification.
- Furthermore, I developed a test GUI using PyQt5 for our glaucoma classification task, enabling input image analysis and delivering conclusive judgments, along with heat maps showcasing the diagnosis.
- Assumed a mentorship role during summer and winter internship periods, providing guidance and support to junior interns within the organization.
- Undertook a notable final project involving the utilization of Generative Adversarial Networks for facial inpainting, aimed at reconstructing partially damaged faces.
Education
May 2022 - May 2024
Master of Data Science
University of Adelaide, Adelaide
GPA:
6.06/7
While pursuing my Masters at the UoA, my primary focus was to further my understanding and expertise in GenAI like Diffusion Models and NLP for generative applications along with managing my coursework. I was also an academic tutor for the following courses:
I also actively interned at:
- Australian Institute for Machine Learning (AIML)
- Centre for Research on Engineering Software Technologies (CREST)
Along with that I also reviewed papers for the following journals:
June 2017 - June 2021
Bachelor of Engineering in Computer Science
University Institute of Engineering & Technology, Panjab University
GPA:
7.69/10
In conjunction with my academic coursework, I actively engaged in projects involving CNNs and GANs. I contributed to the development of tools aimed at facilitating early diagnosis, while also providing mentorship to fellow students in their respective projects.
I also published a few papers during my time at UIET. Some of these are:
- GC-NET for classification of glaucoma in the retinal fundus image
- Fused framework for glaucoma diagnosis using Optical Coherence Tomography (OCT) images
The rest can be found on my Google Scholar page.
Projects
LLM
Code Follow Along
Language Translation
Transformers
Attention Is All You Need
Paper implementation of Attention Is All You Need inspired by Umar Jamil. The model currently supports English to Italian and English to French language translations.
VLM
Code Follow Along
Vision Model
Transformers
PaliGemma: A versatile 3B VLM for transfer
Paper implementation of Paligemma inspired by Umar Jamil. Currently a work in progress
Streamlit
OpenAI
LLM
Langchain
🐱 NekoPDF 📖 - Document Interaction with LLM
A web app to seamlessly interact with your documents using a Large Language Model over the Langchain framework. Currently support OpenAI and a single PDF file at a time but an option for other LLM backends and multiple PDF files will be added soon.
Style Tranfer
Convolutional Neural Network
Style Transfer
Real Time Style Tranfer on videos with Temporal consistency.
Get in Touch
My inbox is always open. Whether you have a question or just want to say hi, I’ll try my best to get back to you!