Testbench

Hardware-agnostic test automation platform. Any UART-capable MCU drops in via a JSON-over-UART protocol. The repo ships five reference firmwares (MSP430FR2355, ESP32, STM32F103C8, RP2040, nRF52840/Zephyr) plus a software-only simulator. A Raspberry Pi bridge streams telemetry over MQTT into a Next.js dashboard with AI root-cause analysis and a built-in troubleshooting chat.

Radiology Second-Opinion Agent

4-person team build (Data & MLOps role). A CV model flags pathologies in chest X-rays, an agentic LangGraph layer cross-references similar cases and literature, and an LLM synthesizes a structured report with ranked diagnoses and confidence scores. Shipped with MLflow model tracking, Evidently drift monitoring, and Kubernetes manifests for deployment.

Backstop  ·  repo

A Zendesk app that measures the hidden workload AI-assisted replies create: the time agents spend verifying, correcting, or overriding AI-generated responses. One-click ticket flagging, per-team verification dashboards, and load-concentration alerts surface a cost most AI rollouts never measure.

Decision Engine

Multi-tenant job scheduler simulator. Takes a resource pool and job queue as JSON and produces a schedule with allocations, utilization/fairness metrics, and a full event log, modeling priority scheduling, deadline tracking, and tenant quotas over labeled resources like a CI runner fleet or GPU lab.

VoiceLegacy

Team entry for Hackabull at USF. A voice preservation tool that clones and archives a person's voice using ElevenLabs, with a simple interface for recording, storing, and replaying.

Edward

Team entry for the IBM Bob hackathon. A dry-witted, voice-enabled desktop AI assistant running locally on Ollama. The concept carried forward into Testbench as its built-in troubleshooting chat interface.