Projects

A production-grade, cloud-native system for high-volume conversion tracking and reporting. Efflux integrates directly with Facebook's Conversion API to stream validated ad conversion data in real time. Designed for scalability, it leverages AWS (S3, SQS, EC2) and databases like MongoDB, ClickHouse, and PostgreSQL to ensure reliable, performant, and fully auditable reporting pipelines—complete with advanced filtering, batching, retries, and monitoring.

A lightweight local ALPR system for macOS that detects vehicles and estimates license plate regions using YOLOv8n, then recognizes plate text with PaddleOCR. Featuring a modern FastAPI backend and web-based UI, it allows users to upload images, process them locally, and visualize results with bounding boxes and confidence scores. While the current model uses a general YOLOv8n detector, a fine-tuned license plate model and full training pipeline are included for future improvements.

This project combines the reasoning power of SMT solvers (Z3, cvc5) with the flexibility of large language models to analyze natural language questions against policy documents. It translates plain English queries into formal constraints, checks them for compliance, and provides clear explanations—bridging the gap between AI-driven understanding and formal verification.

Health Platform – Next.js App. A modern, full-stack health platform built with Next.js. It features secure authentication, multi-step health onboarding, patient management, and Stripe-powered subscriptions. Built using PostgreSQL, Prisma, and Auth0, it's optimized for responsive design and scalable deployment via Docker.

A curated set of machine learning and natural language processing projects, each showcasing standalone implementations of real-world techniques—from DistilBERT fine-tuning and RAG pipelines to Whisper-based audio deepfake detection and foundational ML algorithms like MLPs and gradient descent. Designed for modularity, experimentation, and educational insight, this collection is ideal for developers looking to adapt proven methods into their own pipelines.