About

I am an undergraduate student studying Computer Science at Georgia Tech. I love building things (digital and physical), playing the tabla and teaching.

I am a Research Assistant at the GTRI ARCAID Lab, where I'm working on distributed hierarchical representation learning and synthetic data generation. At the GT Systems for AI Lab, I am developing elastic inference techniques for vision models.

I also occasionally post YouTube videos about cool concepts.

GitHub LinkedIn Google Scholar Email

Experience

    • Developed a comprehensive fraud detection system to flag the use of stolen credentials on Ramp’s bank linking service, saving ~$300k/yr.
    • Overhauled the D1 serverless SQL database dashboard to allow data exploration, ease onboarding & make billing transparent, improving user engagement by 20%.
    • Increased compliance & security by writing an API proxy on Workers (V8 isolates) to intercept user actions and record audit logs using Typescript with Go and Rust.
    • Upgraded D1 insights on the open source “Wrangler” command line tool.
    • Designed & developed a custom evaluation framework for LLM workflows, with a React.js visualization dashboard for clean, extendable & deferrable evaluations.
    • Created novel non-LLM reliant metrics for rapid, zero cost evaluations of LLM-generated code.

  • GT CS 4650: Natural Language Processing

    Jan 2025 - May 2025

    Teaching Assistant

    First undergrad to TA this upper-level course.
Publications

Publications

*= denotes equal contribution

  • KLAS: Using Similarity to Stitch Neural Networks for Improved Accuracy-Efficiency Tradeoffs. Debopam Sanyal, Anantharaman Iyer, Alind Khare, Trisha Jain, Akshay Jajoo, Myungjin Lee, James Clayton Kerce, Alexey Tumanov The Fourteenth International Conference on Learning Representations (ICLR 2026). Paper
  • Enhancing Military Family Readiness and Resilience Programs Through LLM-Generated Synthetic Data and Natural Language Interfaces. Anantharaman Iyer*, Mika Okamoto*, Agustin Lorenzo, Blair Johnson, Moon Kim, Clayton Kerce, Margarita Gonzalez. Presented at Military Health System Research Symposium (MHSRS) in August 2025.
  • Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language models. Aarohi Srivastava et al. (Google Research) Transactions on Machine Learning Research, May 2023. Paper
Research

Research Work

Georgia Tech Research Institute (GTRI) - ARCAID

Research Assistant | January 2024 – Present

Advisor: Clayton Kerce Advisor: Pat Langley
  • Adapting hierarchical representation learning systems to operate in distributed sensor networks.
  • Developed strategies for real-time architecture modifications for efficient model deployment in dynamic compute resource environments.
  • Developed synthetic data generation and retrieval, fine-tuned lightweight classifiers to achieve near parity with Llama-3-70B, enabling field deployment on CPUs.

Georgia Tech - Systems for AI Lab (SAIL)

Research Assistant | January 2025 – Present

Advisor: Alexey Tumanov Website
  • Researching multi-size patch splitting and adaptive patch retention for vision transformers.
  • Modified vision transformer architectures for elastic inference, by patching together parts of different networks using an information similarity selection algorithm. Submitted to ICLR ‘26.

Georgia Tech - NLP X Lab

Research Assistant | January 2024 – May 2024

Advisor: Wei Xu Website
  • Developed a tool to improve information attribution and authorship verification in LLM generated content.
Projects

Projects

  • ShowKASE: Property damage assessment and valuation tool. A tool that combines a damage detection CNN using keras and a price prediction FFNN using scikit-learn to provide real-estate valuation accounting for repair costs. Code
  • Constrained and Optimised Counterfactuals for Interpretable AI. A constrained-optimization method for generating counterfactuals that can be used on a wide variety of datasets and ML models to make those models interpretable and uncover any biases lurking in them. Method demonstrated on several different ML models using MNIST image dataset, Indian Liver dataset and German Credit dataset. Code
  • Mobile app for resolving conflicts among high school students. Created RESCON, the first app to directly engage two people in a conflict through a collaborative game-ified approach.
  • Computational model of the stimulus response of Mimosa pudica. A novel Java-based computational model of the Mimosa pudica leaf that describes its response to touch stimuli of varying intensity by abstracting leaf components as McCulloch-Pitts neurons.
Honors

Honors

  • Dean’s List & Faculty Honors Awarded highest academic standing at Georgia Tech in Fall ‘23, Spring ‘24, Fall ‘24 and Spring ‘25.
  • TMLR Certification Finalist One of five finalists for TMLR’s Outstanding Certification for "Beyond the Imitation Game: Quantifying and extrapolating the capabilities of language models".
  • Linguistics Olympiad Represented India at the Asia Pacific Linguistics Olympiad in ‘19, ‘21 and ‘22.
  • ISEF Finalist Represented India at ISEF in 2019 and 2020.
  • Indian Informatics Olympiad Final round of INOI 2019.
  • CBSE XII Board City Topper Scored 98.4% in the national examination for class XII, topping the city of Bangalore.
  • Best Student of the Year Awarded to one student annually by NPS-HSR Layout “in recognition of hard work, dedication and excellence”.
Contact

Please feel free to reach out, I'd love to hear from you.

📬 aiyer355@gatech.edu
LinkedIn linkedin.com/in/anantharamansi