Graduate Machine Learning Projects

Duration: 4 months

Interpolating Accents in Text-To-Speech

Group Members: Richard Ahn, Shawn Krishnan, Benjamin Lash Course: 10-623 Generative AI

This project presents a continuous-accent TTS model built on Grad-TTS, where accent embeddings are learned and applied via a diffusion framework.

TTS model diagram
  • Implemented a continuous accent embedding network within the Grad-TTS architecture.
  • Preprocessed and labeled Common Voice 12.0 data to enable flexible accent conditioning.
  • Evaluated the model’s performance using objective metrics and a human Mean Opinion Score survey.
Download report

Graduate Mechanical Engineering Projects

Duration: 4 months

CFD Decarbonization Building Study

Group Members: Tej Jolly, Vedant Singh, Shawn Krishnan Course: 24-718 Computational Fluid Dynamics

We investigate CO₂ re-entrainment around building geometries via 2D CFD simulations, identifying optimal aspect ratios that minimize recirculation.

CFD simulation snapshot
  • Configured boundary conditions and solver settings to simulate 2D airflow and CO₂ transport.
  • Ran parametric height-to-width sweeps, finding the 10×20 configuration reduced re-entrainment most effectively.
  • Visualized flow fields and concentration maps to recommend geometry changes for lower turbulence.
Download report
Duration: 4 months

Automated Customizable Ice Dissolving System (A.C.I.D.S)

Group Members: Haoran Zheng, Jianhao Tang, Shawn Krishnan, Haoen Li, Zhaonan Shi Course: 24-671 Electromechanical System Design

This work develops an autonomous turret that dispenses de-icing fluid to melt ice in hard-to-reach areas.

A.C.I.D.S. prototype
  • Conducted user surveys to define requirements and iterated through concept sketches.
  • Prototyped and reinforced the structure, upgrading from servos to stepper motors.
  • Implemented a Raspberry Pi–to–Arduino control pipeline for autonomous targeting and dispensing.
Download report

Undergraduate Mechanical Engineering Projects

Duration: 2 months

Arm-Attachable Vise Assistant

Group Members: Jackson Wood, Shawn Krishnan, Junhan Chen Course: 24-370 Design: Methods & Applications

This design integrates a two-stage gearbox into a wearable brace to reduce wrist torque by 70% when operating a vise.

Vise assistant model
  • Designed a two-stage gearbox based on biomechanical torque analysis to offload user effort.
  • 3D-printed and iterated components after user testing to refine fit and functionality.
  • Incorporated adjustable straps and stops for various hand sizes and torque limits.
Download report
Duration: 4 months

4-DOF Video Jib System

Group Members: Elsa Schleicher, Isaiah Asah, Joshua Pen, Manuel Lancastre, Shawn Krishnan Course: 24-441 Product Design

We built a modular 4-DOF jib, using FEA to optimize stiffness and weight under realistic camera loads.

Video jib diagram
  • Performed FEA on arm segments to determine optimal tube dimensions for minimal deflection.
  • Engineered a backlash-free pan/tilt head using preloaded bearings for smooth motion.
  • Conducted cost analysis to balance material choice between weight and budget.
Download report