Work

Selected engineering work

Selected engineering projects and case studies.

Mobile + Backend

MindMitra

Cognitive-wellness app (iOS + Android)

MindMitra (formerly CognizenX) is a cognitive-wellness app built with DePauw Neuroscience. I led a three-engineer team decoupling AI and data workflows from the shipped React Native client onto a dedicated Vercel/MongoDB backend—isolating ML updates from mobile release cycles while delivering iOS and Android builds with telemetry-driven spaced-repetition scheduling.

Role
Software Engineer – MindMitra
Core challenge
Take over a live v1 product under its prior CognizenX branding, decouple AI from the mobile client, expand to Android, and instrument telemetry for spaced-repetition research workflows without disrupting existing users.
  • Led 3 engineers migrating AI/backend off React Native; shipped iOS + Android updates.
  • 25-endpoint Vercel/MongoDB API centralizing auth, content generation, and scheduling.
  • GPT-4 pipeline: 3,000+ questions across 40+ subdomains, ~300K embedding dedup checks; 65 tests (~60% coverage).
  • React Native
  • Node.js
  • Express
  • MongoDB Atlas
  • GPT-4
  • Vercel

Previously released as CognizenX on the App Store; the product is now branded MindMitra.

Case study
Now available on:

Strong secondary projects

Full-Stack + Applied AI

MediVise

Medical document RAG platform (capstone)

Built a secure, multi-tenant medical document RAG platform with FastAPI, Phi-4-mini, and Supabase/PostgreSQL—enforcing per-user authorization via Row-Level Security and optimizing retrieval with hybrid sparse/dense vectors and cached RxNorm lookups.

Role
Capstone Developer
Core challenge
Deliver a production-quality RAG backend for medical documents within a capstone scope—strict tenant isolation, controlled context windows, and bounded non-diagnostic AI outputs.
  • Multi-tenant Phi-4-mini ingestion API with Supabase RLS on PostgreSQL.
  • Hybrid sparse/dense search, token-aware chunking, and cached RxNorm lookups.
  • FastAPI
  • PostgreSQL
  • Supabase
  • Phi-4-mini

Educational prototype; not intended for diagnosis or medical advice.

Case study

AI Infra + Governance

Fleetrac

AI governance control plane

Building an AI governance platform that ingests OpenTelemetry traces, detects drift and policy violations across simulated agent systems, and drives approval-gated remediation through a FastAPI/SQLAlchemy backend and Next.js operator console.

Role
Independent Developer
Core challenge
Give operators a vendor-neutral way to detect, correlate, and remediate risky agent behavior from live telemetry—not just log dashboards.
  • OpenTelemetry governance platform across 10 simulated agent systems.
  • FastAPI/SQLAlchemy remediation with rolling 15-minute risk windows and approval-gated enforcement.
  • FastAPI
  • Next.js
  • SQLite
  • OpenTelemetry
Case study

Additional featured projects

Full-Stack + RAG

Visa Guardian

Visa-application organization prototype

Web-based AI agent for managing visa applications and tracking progress with intelligent assistance. Built with React, Node.js, PostgreSQL, and a RAG pipeline with Llama 3.1, backed by periodic automated web scraping of official U.S. government sources to ingest the latest visa rules, policy updates, and relevant news.

Role
Developer
Core challenge
Keep AI responses grounded in current regulatory content by automating ingestion of official U.S. government updates while keeping outputs clearly informational.
  • Scheduled web scraping of official U.S. government sites for up-to-date visa rules and policy announcements
  • RAG pipeline with Llama 3.1 for source-grounded, informational assistance
  • React
  • Node.js
  • PostgreSQL
  • RAG

Provides informational workflow assistance and is not legal advice.

Full-Stack + Data

SFO Traffic Tracker

San Francisco traffic visualization

Traffic intelligence platform for San Francisco covering flow, incidents, and road conditions. Built with Next.js and FastAPI, integrating the HERE Technologies API for periodic real-time data ingestion and a cron-based scheduler that throttles polling when no users are on the site to stay within API rate limits.

Role
Developer
Core challenge
Deliver fresh traffic data without exceeding HERE API rate limits by coupling periodic ingestion with visitor-aware polling.
  • HERE Technologies API integration for scheduled, real-time traffic data collection
  • Cron-driven backend that reduces polling frequency during idle periods to preserve API quota
  • Next.js
  • FastAPI
  • HERE Technologies API
  • Cron scheduler

Additional projects

Earlier coursework, tutorials, and exploratory builds.

StudyAI

AI-native iOS study companion for live lecture capture

Built an iOS study companion that turns in-class live notes and voice capture into structured summaries and machine-generated quizzes, powered by Meta's BART seq2seq model on a FastAPI inference layer with offline-first Firestore and Core Data sync.

Anime Recommender

Hybrid recommendation system

Machine learning system using collaborative and content-based filtering with clustering to recommend anime, trained on user and ratings datasets.

MindMatters

Mental health resources web app

Web app supporting mental health journeys with accessible resources, self-reflection tools, and a community space.

DS Tutorial

Customer segmentation tutorial

Data science tutorial for hierarchical clustering in customer segmentation, providing actionable insights for targeted strategies.

Vehicle Repair App

Android repair history tracker

Android app for tracking vehicle ownership and repair history with an intuitive interface, built using Java and XML.

Char Recognizer

Handwritten digit recognition

Java program recognizing handwritten digits through stroke analysis and pattern recognition, demonstrating core computer vision and ML concepts.

Text-Based Adventure Game

Java adventure game

Interactive adventure game inspired by Zork with dynamic navigation, inventory management, and item interaction systems in Java.