Open to opportunities

Hi, I'm Kaavin

Master's student at Rice University (Dec '26). I teach machines to see, learn, and occasionally predict the weather. Currently obsessed with LLMs, MLOps, and making AI that actually works in production.

Machine LearningDeep LearningMLOpsComputer Vision

0

IEEE Papers

0

Internships

0

Certifications

Experience

Teaching Assistant — Automata, Formal Languages & Computability

Rice University

Aug 2025 – PresentHouston, TX

Helping grad students conquer the beautiful chaos of automata theory! I run office hours that actually help (shoutout to the 10+ students whose problem-set scores shot up 📈), craft feedback that makes sense, and keep the course running smoother than a deterministic finite automaton.

TeachingTheoryPython

Software Engineering Intern

VIT Chennai

Jan 2024 – Apr 2024Chennai, India

Built a face-recognition attendance system that knows 500+ faces in under half a second (87% accuracy, 0.5s latency — yes, I timed it). Squeezed 30% more speed out of the inference pipeline, wrapped it in a React Native app, and pitched it so well the university went campus-wide with it. Oh, and it became an IEEE paper. 🎓

Computer VisionReact NativeOpenCV

Machine Learning Research Intern

VIT Chennai

Nov 2023 – Dec 2023Chennai, India

Taught an LSTM to predict wind speeds 10% better than baseline — useful for renewable energy, cooler for my resume. Also got hands-on with federated learning (Flower framework, 4 distributed clients), cutting server load by 35% while keeping everyone's data private. Privacy-preserving ML before it was cool? Maybe. 😎

LSTMFederated LearningFlower

Projects

LLaMA Fine-Tuning for Research Classification

Featured

Spent some quality time with 2k+ arXiv papers so LLaMA didn't have to read them all. Used LoRA to shrink trainable params from 1B → 6M (efficiency win!) and bumped classification accuracy from 40% → 67%. Now it auto-labels papers at ~95ms a pop. 🦙

PyTorchHugging FaceLoRA

MLOps Sentiment Pipeline on AWS

Featured

Built an end-to-end sentiment analyzer that chews through 50k+ IMDB reviews on AWS EKS. Complete with MLflow experiments, DVC for versioning everything (datasets, features, models), and a monitoring stack (Prometheus + Grafana) that would make any DevOps engineer happy. ☁️

AWSDockerMLflowDVC

Federated Learning for Wind Prediction

Privacy-preserving weather predictions! LSTM clients and a Flower server working together without sharing raw data. Renewable energy folks, you're welcome. 🌬️

TensorFlowFlowerLSTM

COVID-19 Detection from Chest X-rays

Custom CNN vs. VGG-16 showdown for medical image classification. Documented all the trade-offs so you can swap in any dataset and get going.

TensorFlowKerasCNN

Face Liveness Detection

Teaching CNNs the difference between 'real human' and 'photo of a human.' Includes data augmentation because fooling this model should at least be a challenge. 👀

TensorFlowOpenCV

Realtime Spam Detection

A logistic regression classifier that hooks into your inbox via IMAP and actually filters spam in real-time. Also logs predictions for drift monitoring — because spam evolves, and so should your model.

scikit-learnTF-IDFIMAP

Skills

Languages

PythonJavaRSQL

ML & Deep Learning

PyTorchTensorFlowKerasscikit-learnHugging FaceLoRALLaMAFlower

Data & Cloud

AWS (EKS, EC2, S3)DockerMLflowPrometheusGrafanaMongoDBDVC

Tools & Frameworks

OpenCVPandasNumPyMatplotlibReact NativeGitCI/CD

Publications

AI-Powered Attendance System Using Facial Recognition

IEEE Conference, 2024

That face-recognition system I built at VIT? It ended up in an IEEE paper! Covers the full pipeline from face encoding to real-time inference at scale. 📸

Enhancing Drug Repositioning Through Collaborative Metric Learning

IEEE, July 2024

A collaborative metric learning approach for predicting drug-disease interactions — turns out the math can outperform traditional repositioning methods on the CTD dataset. 💊

AI Applications in Nutrition & Education

IEEE Conference, 2024

Exploring how AI can make personalized nutrition recommendations and enhance educational outcomes. Because algorithms should help you eat better and learn faster! 🥗📚

Certifications

NPTEL — CSRNPTEL — Educational LeadershipNPTEL — Emotional Intelligence

Designed & built with Next.js, Tailwind CSS, and Framer Motion