I build end-to-end intelligent systems designing RAG pipelines, LLM agents, and agentic tools on the AI side, while architecting the cloud data infrastructure those systems depend on. My work spans AWS, Azure, and GCP, with a focus on retrieval quality, LLM observability, IaC-managed pipelines, and production reliability.
RAG Parent-Child Architecture Arize Phoenix Clinical QA LLM Observability
A domain-specific RAG system for clinical question answering, built on a hierarchical retrieval architecture with parent-child document mapping. Most RAG prototypes use flat vector search — this project directly addresses why those systems hallucinate in production: context fragmentation.
The system retrieves fine-grained child chunks for precision while injecting the full parent document as context into the LLM prompt. Arize Phoenix provides end-to-end LLM tracing and retrieval quality monitoring throughout the pipeline.
Qdrant Sentence Transformers Notion API RAG Pipeline Streamlit
An end-to-end RAG pipeline that indexes entire Notion workspaces and enables natural language Q&A across thousands of pages. The fully automated pipeline handles extraction → chunking → embedding → retrieval → LLM answer generation, with a Streamlit interface built for non-technical users.
LangChain Text-to-SQL LLM Agents Multi-table Schema Streamlit
An agentic system that converts plain-English business questions into accurate SQL, executes queries against relational databases, and returns clean human-readable summaries. Handles multi-table schemas, generates query explanations, and surfaces results in language any stakeholder can act on — removing the SQL barrier entirely.
Terraform AWS Glue Amazon S3 Amazon Athena AWS RDS
Full implementation of the data engineering lifecycle on AWS, built entirely with infrastructure-as-code. Extracts retail data from an RDS OLTP database, transforms it via AWS Glue into columnar Parquet on S3, and serves analytical queries through Athena. Every resource — IAM roles, Glue jobs, S3 buckets, RDS instances — provisioned and version-controlled with Terraform.
Apache Iceberg AWS Lake Formation Amazon Athena RBAC Amazon S3
A governed data lakehouse on AWS using Apache Iceberg as the open table format and Lake Formation for fine-grained role-based access control. Supports ACID transactions, schema evolution, and time-travel queries — production-grade patterns for regulated or multi-team data environments where governance is non-negotiable.
Azure Data Factory Databricks PySpark Synapse Analytics Power BI
End-to-end analytical pipeline spanning ADF for orchestration, Databricks with PySpark for transformation, and Synapse Analytics for serving. Implements Bronze-Silver-Gold medallion architecture — raw ingestion through OLTP-to-OLAP transformation to analytics-ready Gold layer consumed by Power BI dashboards.
PrepWise — AI Interview Simulator · View Repo
A Chrome extension that auto-extracts job descriptions from LinkedIn and generates role-specific interview questions across technical, behavioral, and HR categories. Whisper handles speech-to-text; GPT-4o evaluates answers — the complete interview loop runs in-browser.
GPT-4o OpenAI Whisper Chrome Extension LinkedIn Integration
Languages & Core
AI / ML Frameworks
RAG, Vector Databases & Observability
Cloud & Data Engineering
Infrastructure & DevOps
Databases
APIs & Web
