Explore my portfolio of machine learning, AI, and data engineering projects
19 projects
Fine-tuned an instruction-tuned FLAN-T5 summarization model with Reinforcement Learning from Human Feedback (RLHF) using PPO and a hate-speech reward model to reduce toxicity in generated dialogue summaries.
Designed an end-to-end data platform for a music streaming company (DeFtunes), ingesting purchase events from APIs and a PostgreSQL catalog into an AWS data lake, transforming them with Glue/Spark into Iceberg tables, modeling a star schema in Redshift with dbt, and orchestrating daily runs and data-quality checks with Airflow.
Designed and implemented two complementary ingestion pipelines: a batch/API pipeline that pulls music metadata from the Spotify REST API with authenticated, paginated requests, and a streaming ETL pipeline on AWS Kinesis that processes user activity events in real time and routes them to S3 for downstream recommendation models.
Designed a DataOps-focused project that uses Terraform to provision a secure bastion-host architecture on AWS (VPC, private RDS PostgreSQL, public EC2, SSH keys) and Great Expectations to implement automated data quality checks and documentation on a PostgreSQL dataset.
End-to-end AWS data engineering project designing a NoSQL data layer in DynamoDB and a relational store in RDS PostgreSQL, connected via EC2 bastion hosts and ingesting CSV data from S3 while solving real-world networking, security group, and IAM issues.
Fine-tuned the Flan-T5 language model for dialogue summarization on AWS SageMaker using two strategies: full instruction fine-tuning and parameter-efficient fine-tuning (PEFT) with LoRA. Compared model quality with ROUGE metrics and demonstrated how LoRA adapters can cut model size and compute cost while retaining near-full-tuning performance.
End-to-end data pipeline processing European electricity grid data using cloud-native architecture and modern data engineering practices.
Built a dialogue summarization pipeline using Hugging Face’s FLAN-T5 model, experimenting with zero-shot, one-shot, and few-shot prompt engineering on the DialogueSum dataset to improve summaries of multi-turn conversations
Explore this comprehensive data science project with detailed analysis and implementation.
Explore this comprehensive data science project with detailed analysis and implementation.
Explore this comprehensive data science project with detailed analysis and implementation.
Explore this comprehensive data science project with detailed analysis and implementation.
Explore this comprehensive data science project with detailed analysis and implementation.
Explore this comprehensive data science project with detailed analysis and implementation.
Explore this comprehensive data science project with detailed analysis and implementation.
Explore this comprehensive data science project with detailed analysis and implementation.
Explore this comprehensive data science project with detailed analysis and implementation.
Explore this comprehensive data science project with detailed analysis and implementation.
Explore this comprehensive data science project with detailed analysis and implementation.
