Teaching Assistant, Automata, Formal Languages & Computability
I run office hours and support coursework in automata and computability. I help students break down hard problems and write clearer solutions.
TeachingTheoryPython
Open to ML engineering roles
VisionMLOpsLLMsAI Systems
I build practical AI products.
I am Kaavin Balasubramanian, a machine learning engineer focused on LLMs, computer vision, and MLOps. I care about clear thinking, reliable systems, and good teamwork.
LLM Fine-TuningMLOpsComputer VisionAWSFederated LearningModel MonitoringApplied MLLLM Fine-TuningMLOpsComputer VisionAWSFederated LearningModel MonitoringApplied ML
Projects
FeaturedGOAL
LLaMA Fine-Tuning for Research Classification
I fine-tuned LLaMA with LoRA on 2k+ arXiv papers to classify research topics. Accuracy improved from 40% to 67% with far fewer trainable parameters.
PyTorchHugging FaceLoRA
FeaturedGOALMLOps Sentiment Pipeline on AWS
I built a sentiment pipeline on AWS EKS for 50k+ IMDB reviews. It uses MLflow, DVC, and monitoring so experiments are easy to reproduce.
AWSDockerMLflowDVC
Federated Learning for Wind Prediction
I built a federated LSTM setup with Flower for wind prediction. Each client trains locally, so raw data stays private.
TensorFlowFlowerLSTM
COVID-19 Detection from Chest X-rays
I compared a custom CNN and VGG-16 for COVID detection from chest X-rays and documented what worked, what did not, and why.
TensorFlowKerasCNN
Face Liveness Detection
I trained CNN models for liveness detection to separate real faces from spoof attacks in varied lighting conditions.
TensorFlowOpenCV
Realtime Spam Detection
I built a real-time spam filter connected to IMAP with live inference and feature logs to track drift over time.
scikit-learnTF-IDFIMAP
Experience
Software Engineering Intern
I built a face-recognition attendance system for 500+ users, brought inference down to about 0.5s, and shipped a React Native app used in production.
Computer VisionReact NativeOpenCV
Machine Learning Research Intern
I worked on LSTM wind forecasting and improved results by 10% over baseline. I also set up federated training to reduce central server load while keeping data private.
LSTMFederated LearningFlower
Technical Stack
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
I published a paper on building and deploying a practical facial-recognition attendance pipeline.
Enhancing Drug Repositioning Through Collaborative Metric Learning
I proposed a collaborative metric-learning method for drug-disease prediction and reported better ranking performance on CTD benchmarks.
AI Applications in Nutrition & Education
I reviewed practical AI use cases in nutrition and education, including deployment and evaluation considerations.
Certifications
NPTEL - CSRNPTEL - Educational LeadershipNPTEL - Emotional Intelligence
Open to machine learning engineering roles.
3
Strong
2
Steady
6+
Strong