Featured Work

Projects

Here are some of the projects I've worked on. More will be added soon!

Serverless RAG System with AWS Bedrock

An end-to-end serverless RAG (Retrieval-Augmented Generation) system demonstrating production-grade GenAI architecture on AWS. Key Features: - Designed an end-to-end serverless RAG system using AWS Bedrock, API Gateway, and Lambda with stateless inference - Implemented Amazon Titan Text Embeddings v2 (1024-dim float vectors) stored in OpenSearch Serverless vector engine, with S3-backed knowledge base ingestion - Frontend built using HTML/CSS/JavaScript and hosted on AWS Amplify, demonstrating a cost-efficient GenAI production architecture

AWS BedrockAWS LambdaAPI GatewayOpenSearch ServerlessAmazon TitanS3AWS AmplifyRAGGenAI

Iris ML Model - MLOps Pipeline

Production-ready MLOps demo: FastAPI Iris classifier, Dockerized and Kubernetes-ready. Full lifecycle from training to cloud deployment with Minikube and AWS ECR/EKS.

PythonFastAPIDockerKubernetesAWS ECRAWS EKSscikit-learnMLOpsMinikube

Housing Labs

MLOps infra platform: K3s single-node cluster on Hetzner with Ingress, Service, Deployment and HPA based on resource utilization; Prometheus and Grafana for observability. ML app with XGBoost (tunable parameters) and Streamlit for housing data exploration. Live demo available.

PythonStreamlitXGBoostK3sPrometheusGrafanaHetznerIngressHPAMLOps

Movie Recommendation

Collaborative filtering with matrix factorisation, user-item latent factors, similarity matrices, and hybrid signals. Full ML pipeline containerised with Docker, deployed on Kubernetes (Service + Ingress), served via Streamlit.

collaborative filteringmatrix factorisationDockerK8sStreamlit

Hugging Face Datasets

Open-source datasets on Hugging Face (sweatSmile): FinanceQA, NEET biology QA, medical symptom triage, Bhagavad Gita Q&A, Indian budget 2025, Marx dataset, and more. Used for fine-tuning, RAG, and NLP research.

Hugging FaceDatasetsNLPRAGFine-tuningOpen Source

Marx-OPT Conversational Model

A fine-tuned language model built on facebook/opt-350m optimized for dialogue-style Q&A on political and philosophical topics. The model was trained on a custom dataset of question-answer pairs focusing on Marxist theory, history, and related political contexts.

PyTorchTransformersHugging FaceTRL