Giridhar Vadhul

Experience

• 

  Tech4Good @ UCSC

  Jan 2022 - Sept 2023

  Funded by the Learning Engineering Internship Program to study peer-based feedback for Annota, an intelligent platform for learning qualitative analysis. Received Best Poster Award at HCOMP Doctoral Consortium 2023, Accepted to ACM IUI 2024.• Wrote extensive Python scripts to evaluate feedback in various simulated course settings.• Supervised two interns working on an algorithm for crowdsourced theme discovery.• Led a part-time team working on analytics from platform logs for use in publication.• Conceptualized and derived a novel topic model for unreliable crowd authors and implemented Gibbs Sampling. Show less • Created vectorized implementations of 'Learning from Crowds' and 'Maximum Likelihood Estimation of Observer Error-Rates Using the EM Algorithm' (https://github.com/girivad/Crowdsourcing) to identify relevant text based on crowdsourced annotations.• Achieved 75% Macro-F1 with implementation of Dawid and Skene's algorithm on private, heavily class-imbalanced dataset, beating 46% Macro-F1 for Majority-Vote.• Prototyped multiple EM and EM-inspired algorithms for the above problem.• Developed and deployed a Google Cloud Function based on the Dawid-Skene Algorithm to provide feedback to classes of nearly 150 students.• Created and used multiple automation scripts to populate student assignments and details in database. Show less

  • Learning Algorithms Research Intern

    Jun 2023 - Sept 2023

  • Undergraduate Researcher

    Jan 2022 - Jun 2023
• 

  JLab @ UCSC

  Jun 2023 - now

  NLP Researcher

  • Conducted literature review of Planning for Language Generation, particularly via Monte-Carlo Tree Search.• Working on downstream performance prediction for novel language model architectures.

• 

  University of California, Santa Cruz

  Oct 2023 - Mar 2024

  Group Tutor

  Tutored for ECE 183/283 - Brain-Inspired Machine Learning and independent study of Spiking Neural Networks.Both times, tutored students in PyTorch and Deep Learning, including with the final project on Spiking Neural Networks.

• 

  Xscape Photonics Inc

  Apr 2024 - Oct 2024

  AI Systems Intern

  • Refined and built analytical model of GPT-3 Inference to predict performance on different GPU architectures.• Created composable Python framework to automate roofline analysis by tracing arbitrary LLM architectures.• Conducted sensitivity analysis for KV Caching, Batch Size, Sequence Length and Memory Bandwidth.• Instrumented and profiled distributed Llama-3 training with NVTX Events, PyTorch Hooks and the Nsight Systems Suite and studied communication overhead.• Developing highly concurrent, event-driven simulation of distributed Deep Learning workloads in Golang. Show less