AI Terminal Agent

Building an AI-powered terminal agent that can reason over development tasks, use tools, work with code, and execute multi-step workflows.

Multi-Agent AI RAG System

Designed and built a multi-agent AI RAG system using Google ADK, FastAPI, and ChromaDB, ingesting 36,000 chunks across 11 structured and unstructured sources with modular, testable components.

• Built an LLM evaluation harness using RAGAS-style metrics and 100+ annotated traces to validate output quality and support responsible-AI review before deployment.
• Instrumented end-to-end observability via LangSmith and exported traces to Google Cloud to monitor reliability and diagnose failures in production.
• Reduced p99 latency 29% to approximately 2.3 seconds while holding per-query cost under $0.0005, balancing performance, reliability, and cost.
• Containerized services with Docker for consistent deployment across environments.
• Invited Speaker, NJ AI Literacy Summit — "Building Smarter Conversations: RAG for Education."

Diagnosing and Mitigating Bias in Large Language Models

Evaluated biases in language models including BERT and GPT-2, examining gender, race, and cultural bias.

Speaker Classification using Transformers

Applied self-attention mechanisms and optimized transformer parameters on a speech dataset featuring 600 speakers, achieving 96.85% categorization accuracy and reducing training time by 20%.

HR Dashboard

Designed and developed an interactive Tableau HR dashboard for workforce metrics including attrition rate by gender and job satisfaction to support data-driven decisions.