Aarish Shah Mohsin

Aarish Shah Mohsin
+91 8218594576 | [email protected] | linkedin.com/in/aarish-sm | https://github.com/aarishshahmohsin
Education
Zakir Husain College of Engineering & Technology, Aligarh Muslim University Aligarh, India
Bachelor of Technology in Artificial Intelligence CPI: 8.86/10.00 Sep. 2022 – Jun. 2026
Experience
Apprentice, ML Summer School Jul. 2024 – Aug. 2024
Amazon Inc. Remote
• Completed training in core machine learning topics, covering foundational algorithms and practical applications.
Rsearch Project Dec. 2024 – Present
Department of Computer Science, AMU (Under Dr. Mohammad Nadeem ) Aligarh, Uttar Pradesh
• Working on identifying and removing gender bias from Audio generation Transformer models.
• Testing various models including Meta’s Audiogen, Stable Audio 1.0 for identifying gender bias.
• Debasing these models online using prompt injection and adjustment.
Research Intern Dec. 2024 – Present
Case Western Reserve University (Under Dr. Vincenzo Liberatore) Remote
• Developed and optimized a wide-reach classification algorithm that balances precision and recall by maximizing the
reach of correctly classified positive samples while adhering to strict precision constraints.
• Formulated and implemented a mixed-integer linear programming (MILP) model to identify optimal hyperplanes
for high-dimensional classification tasks with tunable thresholds for precision and margin violations.
AI Research Intern Nov. 2024 – Present
University of Essex, U.K (Under Dr. Amit K Singh) Remote
• Currently Working on Resource efficient Deployments of Large Language Models(LLMs) on GPUs
Projects
LightRag(GraphRAG) based Quiz Generator | Python, GeminiAPI, GraphRag, Pandas, HuggingFace Datasets
• Used the MedMCQA dataset to provide labeled Question/Answers and Context Data
• Used LightRag(modified GraphRag) to convert contexts to a Knowledge Graph(Entity Relationship Graph)
• Used the Google Gemini LLM Model for both Graph generation and Prompting
• High Level queries define the syllabus of the Graph and Low Level Queries produce indepth conceptual questions
Technical Term Text-To-Speech | Python, HuggingFace Transformers, PyTorch
• Used LLM generated templates to generate the dataset. (API gets converted to Ay-Pee-Eye)
• Fine-tuned the SpeechT5 Model for better pronunciation of Technical Terms
• Used the Adam optimizer and applied Post Training Quantization and L1 Pruning
• Used Scaled Log Spectral Distance as an artificial MOS Score
Research
ICASET-2024 | WordNet-Enriched Text Classification with Compressed Distance based Word Networks Jul. 2024
• Used NCD(Normalized Compression Distance) as edge weights for the Word Network.
• Prepared embeddings from Centrality Features and TF-IDF vectors enriched with Word-Net.
• Applied Logistic Regression and Achieved leading performance in few-shot scenarios, Training and Inferring Time.
Open Source Contributions
Pull Requests Typescript, Node, React, GraphQL, Unit Testing
• Added tests for ActionItemUpdateModal.tsx
• Fixed tests for setImageUploadSize.ts
• Upgraded vitest to 1.2.1
• Action check for branch differences
• Fixed localhost access issue
• Tests for getSort.ts
• Refactored setup.ts and added tests
• Fixed currentUserExists.ts tests
• Removed eslint-disable and added workflow
• Removed jest-enzyme and enzyme-react-adapter-17
• Updated event date/time picker
• Image upload size limit
• Fixed reappearing login screen
• Changed User Login and Register components
• Added filtering functionality
• Tests for Advertisements.tsx
• Test for ForgotPassword.tsx
• Docs: added Deduplication Algorithm definition
• Tests for AdvertisementRegister.tsx
• Test for PromotedPost.tsx
Relevant Coursework
• Principles of Machine Learning • Introduction to Artificial Intelligence
• Statistical Learning • AI Tools & Techniques
• Deep Learning • Artificial Neural Network
• Natural Language Processing • ML System Design