Computer Science · Machine Learning · Quant Trader

Hi, I'm Krish.

I'm an undergraduate at Carnegie Mellon studying Computer Science & Machine Learning. I am excited about the intersection of real-world startup products and multimodal dexterity learning for vision and robotics. Interested in alternative transformer techniques, world model vision architectures, and quantitative research/trading.

View Projects

Snapshot

  • School: Carnegie Mellon University
  • Major: Computer Science
  • ML Focus: Computer Vision, Robotics, Diffusion, NLP, Multimodal Learning, Knowledge Distillation
  • Math Focus: Probability, NT, Combinatorics, Optimization, Game Theory, QT

Education

Carnegie Mellon University

B.S. in Computer Science

Undergraduate
  • Concentration: Machine Learning
  • Minors: Mathematics, Robotics
  • Primary interests: Deep learning, quantitative modeling, and systems for vision, language, and robotics

Relevant Coursework

  • Deep Learning (PhD)
  • Machine Learning (Masters)
  • Machine Learning Systems (Masters)
  • Probability
  • Algorithms
  • Vector Calculus
  • Linear Algebra
  • Discrete Math

Honors

  • USAMO National Qualifier
  • USNCO National Qualifier
  • ARML Top 25 National Ranking

Experience

Machine Learning Reseacher

August 2025 – Present
Carnegie Mellon University - Language Technologies Institute
  • Conducting research under Professor Chenyan Xiong to design vision-language models for predictive vision analysis.
  • Designed autoregressive world models to isolate state-transition probabilities for next-frame forecasting, achieving 98% accuracy via CLS fine-tuning.
  • Optimized multi-headed vision-language models via SFT, achieving 92% accuracy.
  • Built contrastive learning models for similarity analysis.

Member of Technical Staff

May 2025 – July 2025
a37
  • Developed reinforcement learning and LLM agents for startup backed by Greylock VC.
  • Designed novel Markov adpative-iterative retrieval architecture, utilizing hybrid dense-sparse vector spaces to maximize semantic information gain within complex repositories, achieving 98% accuracy with ~2 sec latency.

Machine Learning Researcher

Mar 2024 - Present
University of Oklahoma
  • Conducting research under Professor Sridhar Radhakrishnan to natural language processing algorithms for generative drug discovery.
  • Designed novel domain-aware NLP tokenization algorithm using entropy-guided graph filtering for molecular representations.
  • Improved downstream unsupervised learning by 50% compared to SOTA with 50% compression improvement.
  • Developing Transformer architecture with novel position encoding scheme to improve subtree attribution for toxicity performance.

Machine Learning Engineer

August 2024 - Present
Carnegie Mellon Racing - Autonomous Vehicles
  • Designed path planning machine learning software for autonomous robots using LiDAR data.
  • Built low-level SVM midline detection algorithm in C++, improving execution speed by 30% on embedded systems hardware.
  • Accelerated model performance by 10x via multi-threaded binary search in C.

Projects

Selected projects.

Smart Carpool

Intelligent School Carpooling System

Built YOLOv8 computer vision detection model with 87% accuracy. Integrated with Bi-LSTM vision-language model at low-latency to recognize guardian identity instantaneously. Architected Google Firebase infrastructure for real-time synchronization. Designed intuitive front-end iOS app in Swift. Won National Congressional App Challenge.

  • PyTorch
  • YOLOv8
  • Swift
  • Firebase
View on GitHub

InsurAI

AI Financial Healthcare Platform

Developed gradient-boosted insurance models and CNN-based facial analysis. Packaged into an iOS app with SwiftUI + Apple Maps. Won HackCMU 2024 (HRT track).

  • XGBoost
  • PyTorch
  • SwiftUI
  • Swift
  • Apple Maps API
View on GitHub

ProspectIQ

AI Investor Outreach Platform

Full-stack app using React + Flask to automate investor outreach using Gmail API + OpenAI API. Built JSON-based caching to store profiles and communication history.

  • React
  • Flask
  • Python
  • REST API
  • OpenAI API

Sentinel

AI Financial & Risk Detection Platform

Full-stack autopilot financial platform integrated with Plaid and Supabase to analyze and detect high security risk signals. Built ML-powered transaction risk scoring pipeline to surface suspicious fraud signals. Built end-to-end from data ingestion (Plaid) to decisioning layer (OpenAI API) to persistent storage (Supabase) for real-time transaction ingestion adn anomaly learning.

  • React
  • Flask
  • Python
  • Supabase
  • Next.JS
  • Plaid API
  • OpenAI API

Research

Research directions and work in progress.

Generative Models for Drug Discovery

Machine Learning · Generative Modeling

Working with Professor Sridhar Radhakrishnan on natural language processing algorithms for generative drug discovery. Co-authored a paper on a novel string tokenization method for molecular representations. Achieved 50% improvement in compression ratio and entropy over SOTA (APE), and improved downstream clustering accuracy by 30%. Link to Paper

Hybrid Transformer Adaptive Architecture for Generative Drug Discovery

Machine Learning · Large Language Models · Generative Modeling

Leading team to design hybrid Transformer-style adaptive architecture for generative models built on small molecular representations for supervised diffused predictions

Supervised Vision-Language Models for NFL Spatiotemporal Representation

Computer Vision · NLP · LLMs

Designing Vision-Language architecture via RL and SFT techniques to optimize autoregressive world models and isolate state-transition probabilities for next-frame forecasting. Building contrastive learning models for similarity analysis under former Microsoft Researcher Professor Chenyan Xiong.

Autonomous Racing & Trajectory Optimization

Reinforcement Learning · Control · Optimization

Developing global race-line planners and robust control policies for autonomous racing. Focusing on efficient search over trajectories in high-speed, high-uncertainty domains.

Publications

Selected papers.

Published

Optimizing SMILES token sequences via trie-based refinement and transition graph filtering

2025

Sridhar Radhakrishnan, Krish Mody, et al.

Introduced a tokenization method for molecular string representations that improved compression ratio and entropy over prior approaches while boosting downstream clustering accuracy.

View publication

Achievements

Some milestones and recognitions.

Academic & Competitions

  • USAMO & USNCO national qualifier
  • ARML Top 25 National Ranking
  • Selected to Jane Street's QT Program

Hackathons

  • Qualified for National Congressional App Challenge
  • Won 1st at HackCMU in Hudson River Trading track

Contact

Best ways to reach me.

I’m always interested in learning about new research directions, ambitious projects, and technically deep teams.

The fastest way to reach me is by email, but I check LinkedIn regularly as well.