Nikhil Mehta Email & Phone Number

California, United States

Menlo Park, California, United States

GenAI Multimodal Foundation Team, Training Llama Models- Post-training, RLHF, DPO

- Areas: Large Language Models, Generative AI, Interactive Learning, Applied NLP, and Social Science: Published Five Total First-Author Papers at Top Conferences like ACL, NAACL, AACL- Developed methods for humans to interact with AI agents, that the agents can learn from. My approachimproves the performance of LLMs (RLHF), Graph Neural Networks (GNNs), and other ML models.- Researched and developed state-of-the-art ML models- Working with Professor Dan Goldwasser

Teaching:CS 252: Systems ProgrammingCS 373: Data Mining and Machine LearningCS 577: Graduate Natural Language Processing

Greater Seattle Area

- Built an interactive learning solution for a multi-modal task. It enables a LLM-based AI agent to incorporate feedback from humans and ask clarifying questions, making the agent end-to-end interactive. Pre-print available.- Performance improves over 35%.

Cambridge, Massachusetts, United States

- Built a novel ML model to predict Amazon product images given user text queries, published at NeurIPS InterNLP workshop.- Improved performance by over 5% on multiple public datasets

Greater Boston

Improved Named Entity Recognition on Alexa utterances by over 10%, paper published at Amazon conference

San Francisco Bay Area

- Created a patented approach to predict a relevant joke for a chat-bot to reply with, given a conversation history.- Developed technique on top of state of the art (STOTA) NLP models to predict the topic of text from a single sentence with no labeled data. Implemented in cloud.

San Francisco Bay Area

- Used Computer Vision and Deep Learning techniques (CNN) to predict various attributes (age, gender, ethnicity, hair style, etc.) about a person from a single image.- Collected data for and predicted attributes using linear models off of CNN features.- Improved age classification Mean Absolute Error from 5.32 to 3.88 years (27% better).- Created an approach to fine-tune an existing model to easily classify new attributes.- Developed a system to visualize and understand any CNN using Network Dissection.

San Francisco Bay Area

- Applied various Machine Learning algorithms to test new factors for authentication- Implemented a system to identify users based on how they pick up their phone usingsignal preprocessing and Dynamic Time Warping.- Experimented with and evaluated techniques such as Power Spectral Density analysisand Support Vector Machines to authenticate based on human resonance frequency.- Use Deep Learning techniques to publish research on the capacity of Binarized NeuralNetworks at the International Conference on Machine Learning (ICML).

San Francisco Bay Area

- Worked directly with the CEO on product development- Worked on the backend server via REST APIs, RSA Security, and Database interactions- Added features to the Chrome Extension including registration flow and handling of states- Implemented multiple features on the iOS app including RSA encryption, OpenCV face recognition, Data collection, registration flow, and app extension

Doctor Of Philosophy - Phd, Computer Science

Master'S Degree, Computer Science

Focus: Machine Learning and Artificial Intelligence

Bachelor'S Degree, Computer Science

Focus: Machine Intelligence and Security