Featured Work

Building intelligent systems that push the boundaries of what's possible

AI Portfolio Agent

Agentic AI That IS Me

An autonomous AI system that embodies my knowledge, skills, and personality—capable of answering questions about my experience, generating presentations, and continuously improving through self-evaluation loops.

The Challenge

Traditional portfolios are static. Recruiters and collaborators can't interact, ask questions, or get personalized insights. I needed a way to make myself available 24/7 with intelligent, context-aware responses.

The Solution

  • Built a multi-agent reasoning system with tool orchestration for presentation generation using PPTX libraries
  • Implemented a self-improvement loop: GPT-4 evaluates responses against a custom rubric (1-5 score), triggering regeneration until quality threshold is met
  • Integrated RAG pipeline with vector database for context retrieval from my resume, projects, and technical knowledge
  • Deployed full-stack system with FastAPI backend, React frontend, and persistent conversation memory
Response Quality

4.2/5 avg

Regeneration Loops

2-4 iterations

Context Accuracy

95%+

PythonLangChainFastAPIReactTypeScriptVector DBGPT-4

More Projects

Neural Generators Suite

Built GANs, VAEs, and Diffusion models from scratch in PyTorch, achieving high-quality image synthesis with custom training workflows and architecture optimizations.

Challenge: Implementing generative models without high-level abstractions required deep understanding of latent-space manipulation, adversarial training dynamics, and noise schedules.

Key Results:

  • 92% classifier agreement on MNIST synthesis
  • 30-40% reduction in diffusion reconstruction error
  • 20% less framework overhead vs baseline implementations
PythonPyTorchComputer VisionDeep Learning

GraphMatch Pro

Distributed graph-processing pipeline on Hadoop/AWS for large-scale similarity and connectivity analysis across millions of nodes and edges.

Challenge: Processing massive graphs required optimizing MapReduce shuffle operations, handling data skew, and managing distributed computation at scale.

Key Results:

  • 40% increase in processing throughput
  • 60% reduction in end-to-end runtime
  • 25% cost reduction with AWS auto-scaling
HadoopScalaAWS EMRMapReduce

Movie Orchestration Service

Full-stack web application with JWT authentication, GraphQL APIs, and Kubernetes deployment for managing movie metadata at scale.

Challenge: Balancing security, performance, and scalability while supporting traffic surges and maintaining high availability across distributed services.

Key Results:

  • 40% improvement in request throughput
  • 35% reduction in data over-fetching
  • 99%+ uptime with 2-3× traffic handling
ReactGolangPostgreSQLGraphQLDockerKubernetes

CloudBite Delivery Platform

Led 4 engineers in building a real-time restaurant delivery system with separate apps for customers, chefs, and drivers, featuring live order orchestration and multi-threaded processing.

Challenge: Coordinating real-time state synchronization across multiple user types while handling concurrent orders and ensuring drivers don't receive duplicate assignments.

Key Results:

  • Real-time order routing to multiple drivers
  • Thread-safe concurrent order management
  • Dynamic UI updates across 3 separate applications
JavaJavaFXMulti-threadingEvent-Driven Architecture

Library Management System

Comprehensive dual-interface system for library operations with client and librarian portals, handling book inventory, borrowing workflows, payment processing, and automated late fee calculations.

Challenge: Designing a robust database schema to handle complex relationships between users, inventory, transactions, and time-based rules while ensuring data integrity.

Key Results:

  • Dynamic inventory tracking with real-time availability
  • Automated late fee calculation and payment processing
  • Role-based access control for clients and librarians
JavaPostgreSQLJDBCDatabase Design