Parishruthi Ganesh

Currently

• Writing since September 2026

  Preparing a CVPR submission on video anomaly benchmarks

  Extending the pre-event separability diagnostic — scoring anomalous videos using only the frames before the annotated onset — beyond the two benchmarks it was first measured on, and characterising how much of a reported gain survives once that shared source of discrimination is removed.

Last updated 26 October 2026.

Research

My research centres on computer vision, with an emphasis on video understanding and what a video model is actually measuring. My primary focus is human activity detection and object detection, including person detection and multi-object tracking across frames, under weak supervision. I work on benchmark validity — whether a reported gain comes from the representation under test or from something else in the pipeline — and I have applied the same question to model evaluation in language models.

• 

  What Do Interaction Representations Actually Measure? Pre-Event Separability in Weakly-Supervised Violence Detection

  Parishruthi Ganesh

  arXiv'26 · Submitted to WACV

  Paper / Arxiv

  • Video Understanding

• 

  Selecting Open-Weight Language Models for Zero-Shot Intent Classification: A Systematic Evaluation of 41 Models

  Parishruthi Ganesh, Gerry Dozier, Cheryl Seals

  arXiv'26 · Submitted to AAAI

  Paper / Arxiv

  • LLM Evaluation

Projects

• 

  Sentinel Memory

  CockroachDB × AWS — Build with Agentic Memory · August 2026

  An incident-response agent whose memory is a transactional database rather than a context window — it remembers consequences, not conversations.

  Code / Demo (unverified link) / Devpost (unverified link) / Video (unverified link)

  • Agentic Memory
  • TypeScript
  • Next.js
  • React 19

• 

  AstraNova

  Build with Gemini XPRIZE · August 2026

  An AI trading copilot for India's options traders — live market data, AI-scored signals, broker execution and a Gemini-powered assistant.

  Code (private) / Demo (unverified link) / Devpost (unverified link) / Video (unverified link)

  • Applied AI & Data
  • Python
  • Google Gemini
  • Angel One API

• 

  AUSME Faculty Expertise RAG System

  2024 – present

  Retrieval-augmented search and summarisation over roughly 25,000 research papers, built to surface faculty expertise and collaboration opportunities.

  • RAG
  • Python
  • LangGraph
  • Retrieval-Augmented Generation

• 

  DEMA — Digital Engineering Data Management Platform

  2024 – present

  A full-stack desktop application for engineering data lifecycle management, implementing the CIM4DE conceptual information model end to end.

  • Applied AI & Data
  • React
  • Electron
  • Vite

• 

  Circular Trade Fraud Detection with Graph Learning

  2024

  Detecting circular trading rings in iron-dealer invoice data using cycle detection, Node2Vec embeddings and density-based clustering.

  Code

  • Applied AI & Data
  • Python
  • Node2Vec
  • DBSCAN

• 

  Multi-Class Sentiment Analysis with BERT

  2025

  Fine-tuning bert-base-uncased on the SMILE annotation dataset, as an end-to-end study of transformer transfer learning.

  Code

  • LLM Evaluation
  • PyTorch
  • Hugging Face Transformers
  • BERT

• 

  Interactive Rocket Simulation for Science Education

  2024

  A Unity simulation letting students change rocket parameters and watch the resulting flight behaviour in real time, across PC and tablet.

  • Unity
  • C#
  • Physics Simulation

Experience

1. January 2024 – Present

   Graduate Research & Teaching Assistant

   Auburn University

   Auburn, Alabama

   • AUSME Faculty Expertise RAG System
   • DEMA — Digital Engineering Data Management Platform
   • Interactive Rocket Simulation for Science Education
2. December 2021 – July 2023

   Software Developer

   Emids Technologies Pvt Ltd

   Bangalore, India

Miscellaneous