High-performance live CV
Engineered a Python + MediaPipe system for real-time conducting analysis at 30+ FPS.
Conducting Tutor.
Undergraduate research turned into a live conducting coach: webcam input, real-time analysis, and feedback on beat clarity, timing, and posture.
Download for Windows · View on GitHub
Overview
Conducting Tutor was developed at TCNJ under the mentorship of Dr. Andrea Salgian to support conducting fundamentals such as beat timing, time signature clarity, and form awareness during solo practice.
Problem
Beginner conductors often practice alone with mirrors or recordings, which makes objective self-correction hard. The system needed to provide useful feedback in real time, not after the session ended.
The hardest constraint was performance: maintain high FPS while running heavy pose analysis live.
Solution
Core workflow: Camera input → live processing → session ending with visual feedback. The application analyzes conducting motion continuously and surfaces beat, sway, mirroring, and posture signals in a way students can act on immediately.
Validation included faculty/collaborator review and conducting-domain feedback from Dr. David Vickerman.
UX / Interface Design
UI concepts were mapped in Figma, then implemented in a Tkinter interface designed around Shneiderman's Eight Golden Rules so users could run analysis workflows quickly without losing conducting focus.
Architecture
Core stages are split into dedicated modules—segment processing, cycle analysis, and graph generation—so experiments can swap algorithms without rewriting the whole stack.
To keep live performance stable, the system uses a performance-aware pipeline that sustains 30+ FPS and low-latency visual feedback on consumer hardware.
Implementation Highlights
Custom ML model
Trained a Keras model on ~5,000 video samples to replace heuristic beat functions, reaching 87% beat accuracy.
Pedagogy-aware interface
Built a data-driven Tkinter UI for end-to-end practice and configurable feedback visualization.
Results
Conducting Tutor earned 2nd Place (Undergraduate Posters) at the Mid-Atlantic SACNAS Conference, was selected as an NCWIT Collegiate Award Finalist, and was presented at ICMC Boston 2025 and TCNJ's Celebration of Computing.
The project has since been packaged as a standalone Windows application using PyInstaller and is publicly available for download with no setup required.
Download
Conducting Tutor is now packaged as a standalone Windows application. No Python installation or setup required — download, run, and start practicing.
The installer is available on GitHub Releases. Windows only for now.
Windows
Packaged with PyInstaller. Download ConductingTutor.exe from the releases page and run directly.
Research Artifacts
The latest poster from Celebration of Student Achievement (COSA) is embedded below. The earlier SACNAS capstone poster PDF opens in a new tab or can be previewed in a popup.
Earlier artifact: SACNAS capstone poster (PDF) ·
Read more: ICMC 2025 paper · full research paper
Milestones from this research journey: SACNAS placement, Celebration of Computing presentation, and ICMC Boston.
Earlier milestones: SACNAS second place · Celebration of Computing · ICMC Boston 2025
Reflection
The most important lesson was balancing research ambition with real-time product constraints. Prioritizing clarity, responsiveness, and measurable feedback made the system more useful for actual student practice.
RicoCibo
A UI/UX case study for a mobile app that combines recipe discovery, smart grocery lists, and expiration tracking.
All projects