Looking for a job

Sai Aditya Patkuri

@saiadityapatkuri

Message

Computer Science at IIT Bombay '24 with a strong interest in Machine Learning.

India

Message

What I'm looking for

I'm passionate about machine learning, especially working with LLMs and exploring their potential. I'd love a role where I can dive deeper into this technology, collaborate with other ML enthusiasts, and apply these models to real-world challenges. I'm constantly learning in this field and want to be somewhere that values innovation and creative problem-solving with AI.

As a B.Tech student in Computer Science and Engineering at the Indian Institute of Technology Bombay, I have consistently demonstrated my passion for technology and innovation. My academic journey is marked by notable achievements, including securing a prestigious KVPY Fellowship and excelling in competitive exams like JEE Advanced and JEE Main. These accomplishments reflect my dedication and ability to thrive in challenging environments.

During my internship as a Unity Developer at Super Huge Studios, I designed and developed engaging levels for mobile games, showcasing my skills in game development and collaboration. My projects, such as the Repository-Level Prompt Generation for LLMs and the Detection of Hate Speech in Memes, highlight my proficiency in machine learning and natural language processing. I am eager to leverage my technical expertise and creativity to contribute to innovative projects in the tech industry.

Experience

Work history, roles, and key accomplishments

Course Project

Indian Institute of Technology Bombay

Nov 2023 - Dec 2023 (1 month)

Repository-Level Prompt Generation for LLMs
Guide: Prof. Balamurugan Palaniappan

• Adapted repository-wide prompt generation frameworks (RLPG, PPC) to boost LLM-based code completion.
• Achieved a 10.9% success rate improvement using StarCoder API and CodeBERT for prompt evaluation.
• Built automation pipelines for AST parsing, hole detection, and prompt proposal generation.

BERT Python LLM Code Completion

Unity Developer Intern

Super Huge Studios

May 2023 - Jul 2023 (2 months)

• Designed and developed 10+ levels for mobile games like SolitaireTriPeaks and BucketGame.
• Collaborated with colleagues and sought mentorship to gain insights into game development.
• Engineered a C# prototype for an upcoming project with interactive levers, used it to test assets, design, and
mechanics.

Unity C++Game Development

Course Project

Indian Institute of Technology Bombay

Mar 2023 - May 2023 (2 months)

Detection of Hate Speech In Memes
Guide: Prof. Dr. Pushpak Bhattacharyya

• Explored integrating Object Detection and Image Captioning to develop a multimodal Hateful Memes classifier.
• Investigated applying Multi-Model Baselines like BERT+ResNet to tackle unimodal model limitations.
• Fine-tuned VisualBERT pre-trained on COCO, optimized it for multimodal, then validated on 500 memes.

Python Object Detection Machine Learning BERT

Course Project

Indian Institute of Technology Bombay

Nov 2022 - Dec 2022 (1 month)

POS Tagger
Guide: Prof. Dr. Pushpak Bhattacharyya

• Implemented a POS tagger based on a Hidden Markov Model, applying the Viterbi algorithm.
• Engineered a POS tagger utilizing Word2Vec vectors coupled with a FFNN-BP Architecture.
• Attained an accuracy of 97% and achieved all performance metrics (Precision, Recall, F1-Score) above 90%.

Python Natural Language Processing Viterbi Algorithm Word2Vec Hidden Markov Models

Course Project

Indian Institute of Technology Bombay

Oct 2022 - Nov 2022 (1 month)

Vector Based Word Sense Disambiguation
Guide: Prof. Dr. Pushpak Bhattacharyya

• Produced noun synset IDs using Extended Lesk and PageRank methods with the SemCor dataset.
• Employed WordNet’s first sense as Extended Lesk baseline, achieving 0.32 value and 38% accuracy.
• Attained 48% accuracy and exceeding 40% performance metrics for the PageRank-based approach.

Python Word Sense Disambiguation PageRank SemCor Extended Lesk

Course Project

Indian Institute of Technology Bombay

Apr 2022 - May 2022 (1 month)

Socket Programming
Guide: Prof. Kameswari Chebrolu

• Implemented a P2P network enabling file search and download up to a specified neighbor depth.
• Leveraged multi-threading and the client-server model for downloading files from peers.
• Utilized C++ to establish TCP connections for file downloads and MD5 hashing to detect errors.

C++Networking Client Server Architecture TCP