Scribd
Upload
44 views2 pages

Aditya Shah CV PDF

Uploaded by

9960dhirajmanje
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
44 views2 pages

Aditya Shah CV PDF

Uploaded by

9960dhirajmanje
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

Aditya Shah

ƒ +1 (540) 824 9021 # aditya.shahh3@gmail.com ï aditya-shahh ³ Google Scholar


Summary
• I am closely working with the leadership and research teams to build large-scale enterprise GenAI solutions.
• My core responsibilities include:
◦ Implementing Large Language Model architectures (Llama2, Mixtral, etc) and performing domain adaptive pre-training on
multi-GPU sytems using DeepSpeed / Megatron.
◦ Synthesizing domain specific enterprise data into required prompt-instruction pairs and performing instruction fine-tuning
using DPO / RLHF.
◦ Developing a solid understanding of novel architectural innovations (KV cache, flash attention, MQA, etc), optimization
techniques (quantization, FSDP, etc), and implementing newer research methods.
• Key Expertise:
◦ Tools/Tech: Python, PyTorch, DeepSpeed, Megatron-LM, Accelerate, GPU, CUDA, Docker, NumPy, Pandas, SQL.
◦ Deep Learning: Pretraining & Finetuning Large Language Models (LLMs), RLHF, DPO, FSDP, RAG, Vector DB.

Work Experience
Capital One Mclean, USA
•
Machine Learning Scientist - AI Foundations Jun 2023 - Present
◦ Built in-house models with Llama2 and Mixtral, loaded checkpoint weights, and conducted further pre-training on
enterprise data (casual language modelling) using Megatron, FSDP on multiple GPUs.
◦ Developed prompt-instruction dataset from enterprise data and fine-tuned these base models using DPO/RLHF to align
them for chat-based use cases.
◦ Implemented various optimization techniques like KV cache, reduced precision, Multi Query Attention, Rotary Position
Embeddings, etc to optimize fine-tuning and inference pipelines.
◦ Building domain specific LLM agents using RAG and VectorDB to provide AI based virtual assistance with different
financial legalities and preventing risks.
◦ Delivered various keynotes and training sessions on Generative AI, and NLP, highlighting personal expertise and
leadership in upskilling teams.

Google Seattle, USA


•
Research Scientist Intern - LLMs Sep 2022 - Dec 2022
◦ Worked with DeepMind to integrate soft prompt parameters and adapters in a Multimodal Large Language Model
(MLLM) for Document Extraction.
◦ Developed an efficient optimization pipeline and performed parameter-efficient prefix fine-tuning on TPUs to
extract data from invoice documents.
◦ Enhanced model’s adaptability and robustness in sequential uptraining, which reduced catastrophic forgetting by 14%.

Capital One Mclean, USA


•
Data Science Intern - NLP Jun 2022 - Aug 2022
◦ Fine-tuned transformer-based language models (RoBERTa, XLNet, T5) on enterprise wide call transcript data to extract
relevant knowledge, identify entities and summarize the transcript.
◦ Improved customer request fulfillment and agent performance through co-reference resolution and eliminated 70% of
false positives with 94% accuracy.

Indian Institute of Technology (IIT) Indore, India


•
Research Scientist - Machine Learning Sep 2020 - Aug 2021
◦ Developed a novel multimodal neural network architecture for sarcasm detection which outperformed existing benchmarks
by 6.14% F1 score. Research Paper accepted in ICONIP 2021
◦ Proposed an efficient self-attention based model to capture incongruity for code-mixed sarcasm detection. Achieved
competitive F1 score as compared to multilingual models while training 10x faster and using lower memory footprint.
Research Paper accepted in ICON — (ACL 2021)

Saarthi.ai Bangalore, India


•
Machine Learning Engineer Jul 2020 - Oct 2020
◦ Conducted applied research on ASR and developed a deep learning model based on BiLSTM and 1-D CNN for gender
identification from audio data.
◦ Achieved a test accuracy of 96% with 15% improvement over previously designed approach. Further, worked on age
identification and specific keyword detection from real-time audio input.
Selected Publications
• A. Shah, A. Jain, S. Thapa, and L. Huang, “ADEPT: Adapter-based Efficient Prompt Tuning Approach for Language
Models”, The 61st Annual Meeting of the Association for Computational Linguistics (ACL), 2023 Paper

• B. Yao*, A. Shah*, L. Sun, and L. Huang, “End-to-End Multimodal Fact-Checking and Explanation Generation: A
Challenging Dataset and Models”, International ACM SIGIR Conference on Research and Development in
Information Retrieval (SIGIR), 2023 Paper (Best Paper Honorable Mention)

• S. Thapa*, A. Shah*, F. Jafri, U. Naseem, and I. Razzak, “A Multi-Modal Dataset for Hate Speech Detection on Social
Media:Case-study of Russia-Ukraine Conflict”, Conference on Empirical Methods in Natural Language Processing
(EMNLP), 2022. Paper

• A. Shah and C. Maurya, “How effective is incongruity? Implications for code-mixed sarcasm detection”, Proceedings of
the 18th International Conference on Natural Language Processing — (ACL), 2021. Code Paper

• S. Gupta, A. Shah, M. Shah, L. Syiemlieh, and C. Maurya, “FiLMing Multimodal Sarcasm Detection with Attention”,
Proceedings of the 28th International Conference on Neural Information Processing (ICONIP), 2021. Code Paper

• L. Kurup, M. Narvekar, R. Sarvaiya, and A. Shah, “Evolution of Neural Text Generation: A Comparative Analysis”,
Advances in Intelligent Systems and Computing, Springer (IC4S), 2020. Paper

Skills Summary
• Libraries and Technologies: PyTorch, DeepSpeed, Megatron-LM, Accelerate, GPU, CUDA, NumPy, Pandas, SpaCy
• Languages & Frameworks: Python, C++, SQL, MongoDB, Flask, Docker, Kubernetes, Spark

Education
Virginia Tech Blacksburg, USA
•
Masters of Science in Computer Science - Research 2021 - 2023
Thesis: NLP based Episodic Future Thinking (EFT). (Funded by NIH)

Dwarkadas J. Sanghvi College of Engineering Mumbai, India


•
Bachelors of Science in Computer Science 2016 - 2020

Academic Projects
• Code Interpretability on transformer models using SHAP: Conducted an exclusive research study on analysing code
interpretability using SHAP values and Logit manipulation for Codebert and Graph Codebert models. Code

• Adaptive pooling based Electra model for Multi Label Relation Classification: Proposed an Adaptive pooling based
method on top of Electra model — AdaElectra for multilabel relation classification achieving F1 score of 0.88 on the NYT29
dataset. Code

• Weighted Contextual N-gram method for evaluation of Text Summarization: Finetuned T5 model on Extreme
Summarization (XSum) Dataset and proposed the use of Weighted Contextual N-gram (WCN) method – an alternative metric for
evaluation of text generation. Code

• Supervised Text Generation using GPT2 model, BiLSTM, and GloVe Embedding: Fined tuned GPT2 model on
wikisent data for generating context-dependent text samples. Developed a BiLSTM with GloVe embedding and N-gram model to
generate text with 90% test accuracy. Code

• Food-101 Challenge by ETH Zurich: Designed a Neural Network model on top of the Xception network and fine-tuned it to
achieve State-of-the-Art result on the challenging Food 101 Dataset with a test accuracy of 87%. Code

Honors and Awards


• Received “Best Paper Honorable Mention” for the work on Multimodal Fact Checking at ’SIGIR’, 2023.
• Served as a Reviewer for: NAACL SRW 2023, ICON 2023, EMNLP 2022, COLING 2022, ICON - ACL 2021.
• Selected for AI fellowship program, Fellowship.ai, May 2020.
• Awarded “Best Research Project” at HaXplore, IIT BHU Machine Learning hackathon, 2019.
• Received “Innovative Research Project” award in ‘CodeShastra Intercollege Hackathon’, 2019.
• Served as “Co-Technical Head” for ACM, 2017-18. Mentored a team of 10 students for Software Development and ML.
• Awarded “Google India Scholarship”, 2017. in Android Application Development.

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy