Pravir Chugh Email & Phone Number

Palo Alto, California, United States

• Architected and deployed an AI-powered code summarization tool using LLMs, reducing code review time from five days to two days (60% time saved) for 50+ engineers on the Buying and Contracts teams.• Implemented threaded processes and asynchronous programming to optimize internal LLM infrastructure, increasing throughput from 15 to 100 API calls per minute, enhancing performance and scalability for Ariba developers.• Enhanced debugging efficiency using a novel application of OpenAI and Gemini APIs and RAG to analyze stack traces, addressing scaling challenges of searching across large codebases within a limited context window.

Los Angeles, California, United States

• Secured partnerships with AI industry leaders, resulting in internship opportunities for club members.• Delivered a comprehensive talk on the working mechanisms and fine-tuning processes behind large language models. • Created a CNN-based model to identify food items in refrigerators, paired with a recipe generation algorithm for healthier meal planning.

• Spearheaded the development of a profile sorting algorithm to extend the capabilities of Canopy, a cross-platform application designed to forge peer networks and mentorship opportunities for over 2K college students.• Created GraphQL queries and used TypeScript hooks to streamline profile list rendering, enhancing UI consistency.• Presented proposals for development initiatives to project leaders and key stakeholders representing various academic institutions, including UCLA and the University of Chicago.

• Optimized a Llama 2 model with targeted dialogues to power a motivational LLM coach for a personalized fitness app, enhancing user engagement, experience, and safety.• Deployed and integrated the AI model into the app backend using TypeScript.

• Fine-tuned BERT models using TensorFlow and Transformers, uncovering a 15% increase in bias in generative LLMs through word embedding analysis. • Leveraged GPT-4 and PaLM APIs to categorize 100K+ LLM-generated outputs for large-scale data processing.• Pioneered a novel methodology to detect and quantify bias in both proprietary and open-source generative models, adapting embeddings-based approaches like WEAT (Caliskan, Bryson, & Narayanan, 2017) to the generative space.

• Web Developer for ACM at UCLA's online code editor, designed to help introduce students in the LA area to different facets of computer science, including web development, Python, graphics libraries, and other topics. • Added bracket completion capability and optimized output and terminal viewing windows for users.

Bachelor Of Science - Bs, Computer Science

High School Diploma

Summer Institute For Mathematics

One of the 30 students selected to participate in the summer program. Studies in SIMUW 2021 focused on fields including proofs, group.