Problem:

Make compact and affordable swerve drivetrain modules for FTC robots, since traditional pods are too big and too expensive for most teams.

Method:

• Designed first prototype under 4in × 4in × 4in using 3D-printed structural parts to hold shafts, gears, and bearings.

• Manufactured and assembled a full drivetrain using four pods (1 motor + 1 servo per pod).

• Evaluated costs and mechanical performance.

• Created a second prototype with changes:

  • Switched to generic parts from Amazon/AliExpress to reduce cost.

  • Increased module height by < ¼ inch.

  • Changed module shape from L-shaped to rectangle.

  • Updated hole pattern to fit Gobilda, REV, and Andymark build systems.

Result:

• First prototype worked mechanically but was too expensive (~$350 per pod, ~$1300 for a full chassis).

• Second prototype reduced cost and improved compatibility, making it more realistic for teams to use.

• Project is ongoing — aim is to create a drivetrain that is both affordable and competitive for FTC teams worldwide.

Click the photo to find out more!

FTC Coaxial Swerve

Click the photo to find out more!

Problem:

After losing the 2023-2024 New York City FTC Championship, our team wanted redemption in an off-season event. We decided to build an entirely new swerve-drive robot—a drivetrain rarely seen in FTC because of its mechanical and programming complexity. Swerve drives retain traction while still offering omnidirectional motion, unlike mecanum or omni-wheel drives that lose grip because of their rollers.

Method:

• Simple Swerve Prototype:

  • Studied differential swerve mechanics (two drive motors per wheel controlling both rotation + steering).

  • Referenced FTC 11115 Gluten-Free’s design and simplified it for cost and accessibility.

  • Re-engineered gearbox: replaced 3D-printed D-bore gears and set-screws with M3 × 45 mm bolts + bearings embedded in the gears for stronger alignment.

  • Result: cheaper, easier-to-replicate pod while preserving drivetrain geometry.

• Simple Swerve V1:

  • Realized full pod stack was 18.25 in long, exceeding FTC 18 in limit.

  • Rotated motors 90° with REV Ultra 90-degree gearboxes to shorten module length.

  • Achieved FTC-legal chassis footprint.

  • Downside: added extra bevel gear pair, slightly reducing efficiency.

• Simple Swerve Chassis:

  • Built 2-pod chassis with four corner omni-wheels for stability.

  • Added cutouts for linear slides to compete in an off-season event.

  • Event was canceled before testing, but mechanical assembly completed.

• Simple Swerve V2:

  • Further reduced bill of materials and offered two motor configurations:

    • REV Ultraplanetary Motors – compact form factor, no external gearbox.

    • GoBilda Yellowjacket Motors – 8 mm REX shafts for durability.

  • Both configurations reuse the same internals; only the casing and outer plates differ.

  • Final version achieved smaller, cheaper, and easier-to-manufacture modules.

Result:

• Completed design of a cost-efficient differential swerve module for FTC use.

• Demonstrated that a reliable swerve drivetrain can be built within FTC size constraints and on a reasonable budget.

• Although the chassis never competed, the project produced a working prototype that can serve as a foundation for future FTC swerve robots.

Simple Swerve

Problem:

Build a lifelike, voice-activated AI assistant that can recognize speech, respond intelligently, and display natural eye movements with minimal noise interference.

Method:

• Eyeball mechanism:

  • Designed the eye so servos are hidden inside.

  • One servo pivots the eye up/down; a second rotates the first mount for left/right motion.

  • Entire system mounted to a bolt on a spare monitor arm.

• Upper eyelid:

  • Located mounting holes on the arm and designed a pivot mount.

  • Connected servo horn to eyelid via linkage.

  • Reinforced the mount after initial design broke under stress.

• Both eyelids:

  • Lower eyelid pivot point made 5 mm thicker for durability.

  • Added slot for upper eyelid to slide into (both pivot on same axis).

  • Used longer linkage to account for servo placement.

• Software integration:

  • Connected wake-word detection, speech recognition, and ChatGPT for conversation.

  • Implemented cosine-based easing for smooth servo acceleration/deceleration.

  • Programmed synchronized blinking and idle/listening/speaking states.

  • Ensured servos pause during listening to reduce microphone noise.

Result:

Created a finished AI assistant with:

• Real-time speech recognition and ChatGPT responses.

• Smooth, lifelike eye motion and synchronized eyelid blinking.

• Ability to wake on command, hold a conversation, be interrupted mid-speech, and reset automatically.

• The system successfully blends mechanical realism with intelligent voice interaction.

Click the photo to find out more!

A-Eye Animatronic Chatbot

FTC Team 9384 2023 - 2024 Robot "EggWUUUHH"

Click the photo to find out more!

Problem:

Design and build a competitive FTC robot for the 2023–2024 CENTERSTAGE game, capable of scoring hexagonal pixels on a tilted backdrop, completing side-objectives like plane launching and climbing, and iterating quickly across the build season to improve performance.

Method:

• September 9 – December 9 (Kickoff → Qualifier 2):

  • Linear Slides: Built stacked drawer-slide system with 3D printed spacers/pulleys to reach all scoring heights.

  • Chassis: Custom aluminum sheet chassis with mecanum drive; 4 plates parallel with cutouts and mounting holes.

  • Intake: 3-stage active intake (star wheels → counter rollers → zip-tie stage) for fast dual-pixel capture.

  • Outtake: Pixel box with 3D-printed rubber belts; struggled to score at correct angle.

  • Plane Launcher: Spring-powered mechanism with two servos (release + angle control).

• December 9 – January 13 (Qualifier 2 → Qualifier 4):

• Slides/Intake: Unchanged.

• Outtake: Redesigned to pivot down and place pixels directly onto backdrop; added pinching mechanism for accuracy.

• Plane Launcher/Chassis: Unchanged.

• January 13 (Qualifier 4 → Super Qualifier):

  • Slides/Chassis/Outtake: Unchanged.

  • Intake: Added servo-powered pincher for autonomous pre-stacked pixels.

  • Plane Launcher: Same mechanism, but frame improved aesthetically.

• Super Qualifier → State Championship:

  • Slides/Chassis/Intake/Outtake/Launcher: Unchanged.

  • Climber: Finalized a winch-based climber with pivoting hooks, string actuation, and passive rubber-band assist for fast, strong climbs.

Result:

• Built four major robot iterations over the season.

• Final robot included reliable linear slides, mecanum chassis, dual-intake, accurate backdrop outtake, spring plane launcher, and fast winch climber.

• Achieved a robust, competitive design that evolved in response to weaknesses of earlier prototypes.

• Engineering portfolio and CAD were finalized and published (GrabCAD + GitHub).

Problem:

Build a robot for the 2022–2023 FTC game POWERPLAY that can reliably pick up and place cones onto poles of varying heights, while being stable and efficient across multiple matches.

Method:

• September 10 – December 4 (Kickoff → Qualifier 4):

  • Linear Slides: Used Rev Robotics linear slide kit for reach and simplicity. Intake mounted to bottom of final stage. Effective but swayed at max extension, causing cone drops.

  • Chassis: Simple mecanum drive on C-channels arranged in an H-pattern. Motors bolted with L-brackets. Design based on Rev Robotics Mecanum Kit.

  • Intake: Dual compliant wheels on servos grip cone by pinching it in the center. Worked, but struggled to keep cone secure.

• December 4 – February 12 (Qualifier 4 → Super Qualifier):

  • Linear Slides: Same slides, but reduced gear ratio for faster lift.

  • Chassis: Unchanged.

  • Intake: Replaced wheel intake with claw-based design for reliable cone holding.

  • Turret: Added direct-drive turret with lazy susan bearing so robot could score on poles without turning the chassis. Shaft passed through lazy susan center — unique for FTC at the time.

• February 12 – State Championship (Super Qualifier → States):

  • Linear Slides: Added second slider for extra rigidity, eliminating sway at max extension.

  • Chassis: Redesigned with bevel gearboxes so motors sat inside C-channels, lowering chassis and squaring the frame. Reinforced with extra cross beams to support larger turret.

  • Intake: Same claw system.

  • Turret: Upgraded to larger lazy susan for more surface area, allowing space for dual linear slides.
    

Result:

• Iterated through three major designs.

• Final robot had dual rigid slides, low-profile reinforced mecanum chassis, reliable claw intake, and strong turret system.

• Addressed key weaknesses: stability, speed, and alignment when scoring.

• Finished build season with a polished, competitive machine, documented with CAD and notebook.

Click the photo to find out more!

FTC Team 9384 2022 - 2023 Robot "Crabby"

3D Printers

Click the photos to find out more!

Delta 3D Printer

• Effector V3:

  • Re-mounted hotend lower to reduce center of gravity.

  • Extruder placed just above hotend.

  • Redesigned ducts to concentrate airflow at nozzle tip.

  • Switched to detachable membrane-switch probe mounted directly on nozzle for accurate calibration.

• Closed-Loop Stepper Motors:

  • Added encoders to stepper shafts → servo-like behavior.

  • Eliminated step loss, reduced resonance, smoother motion.

  • Used drivers mimicking A4988 to communicate with motherboard.

• Touchscreen Interface:

  • Added TFT35 touchscreen running Klipperscreen.

  • Mimics full web interface, mounted on pivoting bracket above electronics box.

    
    

Result:

• Built a finished delta 3D printer capable of very fast and consistent prints.

• Closed-loop steppers provided reliability and accuracy.

• CPAP-based cooling and direct-drive effector solved weight and extrusion issues.

• Iterative effector designs improved calibration accuracy, stability, and part cooling.

• End product is a unique, high-performance delta system with a custom interface and cutting-edge motion control.

Problem:

Design a high-speed, high-quality delta-style 3D printer using closed-loop steppers, a lightweight effector, and extremely strong cooling, while overcoming the complexity of delta kinematics and the challenges of weight, extrusion consistency, and airflow.

Method:

• Base System:

  • Reused frame, PSU, and SSR bed heater from a TEVO Little Monster.

  • Added custom effector, closed-loop Nema 17 steppers, TFT touchscreen, and belt tensioners.

• Effector V1:

  • Tested Bowden extruder to minimize weight.

  • Found long filament path caused friction → inconsistent extrusion + print artifacts.

  • Fan ducts failed to deliver stable airflow.

  • Outcome: decided to move to direct drive extruder + stronger cooling solution.

• CPAP Fan Cooling:

  • Mounted CPAP blower to printer frame.

  • Routed airflow via hose to ducts at effector.

  • Reduced effector weight and opened up more space for hotend/extruder.

• Effector V2:

  • Fully integrated design: ducts, hotend mount, extruder mount, bed probe mount in one part.

  • Added direct drive extruder for consistent extrusion.

  • Upgraded hotend for higher flow rate + larger melt zone.

  • CPAP fan cooling integrated into hotend.

  • Issues: probe was offset → inconsistent leveling, ducts failed to concentrate airflow.

Click the photo to find out more!

ZeroG Hydra

Result:

• Still under development, but design already integrates:

  • Large build volume with CoreXY motion for balanced speed and accuracy.

  • Fully automated calibration through floating bed + Beacon probe.

  • Advanced toolhead sensors for reliable filament handling and multi-material support.

  • High-performance hotend/extruder combo ensuring consistent extrusion and top-tier print quality.

• Aims to become a flagship quality-focused printer in your collection, complementing your other high-speed builds.

Problem:

After building multiple printers—both custom and off-the-shelf—I needed a large build volume machine that could reliably produce high-quality prints. Speed was not the focus here; the goal was a printer with advanced sensors and robust hardware to consistently output precise, strong parts.

Method:

• CoreXY Motion System:

  • Chose CoreXY for fast, precise motion with minimal moving mass.

  • Two stationary motors drive X and Y axes through a belt system.

  • Results in lighter gantry and smoother high-speed movement.

• “Floating” Bed Assembly:

  • Designed a 3-point, independent pivoting bed mount.

  • Allows automatic tramming and firmware-controlled bed leveling.

  • Essential for fully automated calibration before each print.

• Electronics Bay:

  • Placed under the printer for compact design.

  • Includes: USB splitter, mainboard, PSU, SSR for bed heater.

  • SSR enables 110V bed heating for large build plate.

  • Runs Klipper firmware on Linux for flexible, high-performance control.

• Modernized Toolhead:

  • Based on Filimetrix, with extensive sensor integration.

  • Features:

    • Dual filament monitoring sensors

    • Filament cutter knife

    • Eddy current bed leveling probe (Beacon)

    • LED indicators + touchscreen

    • Phaetus Rapido hotend (high flow, rapid heating)

    • Clockwork 2 extruder (precise filament control)

  • Improves usability and diagnostic feedback.

• Toolhead Sensors:

  • Filament monitoring: Two limit switches track runout, feeding, and cutting.

    • Top sensor → runout + feeding detection.

    • Bottom sensor → confirms cut for multi-material workflows.

  • Beacon probe:

    • Magnetic/eddy current bed leveling with accelerometer function.

    • Provides fast, accurate meshes and nozzle-based Z-offset calibration.

Click the photo to find out more!

Voron V0.2990

Problem:

I wanted to build a Voron printer because of the community-driven ethos: there are no official kits, only BOMs, CAD, and manuals, which means each build is unique. After completing my machine, I quickly noticed several bottlenecks and began designing modifications to push performance further.

Method:

• Custom Toolhead:

  • Stock Voron extruder/hotend couldn’t keep up with very high speeds (225–350 mm/s printing, 650 mm/s travel).

  • Designed my own toolhead with:

    • More powerful extruder (higher current, better grip with higher gear ratio).

    • Hotend with larger melt zone and optimized nozzle for higher flow rate.

  • Entire toolhead frame 3D printed.

  • Result: maintained extrusion at speeds higher than motion system’s practical limits.

• Auxiliary Cooling Fan:

  • External 12032 grill fan for additional part cooling without adding gantry weight.

• Active Carbon Filter:

  • 5015 blower pushes chamber air through carbon filter.

  • Removes VOCs from toxic filament printing.

• Chamber Temperature Sensor:

  • Reused a spare thermistor.

  • Mounted at top of internal frame for accurate chamber temp readings.

• Camera Mount & AI Integration:

  • Added camera feed to web interface.

  • Uses open-source script for recording, timelapses, and auto-pausing failed prints.

• Filament Runout Sensor:

  • Encoder + limit switch to detect filament presence and extrusion rate.

  • Mounted on back panel above electronics bay.

• Semi-Custom RGB Panels & Hinges:

  • Replaced failed 3D-printed hinges with reinforced design using longer heat-set inserts.

  • Added protective door clips and smoked acrylic panel with RGB diffuser for camera visibility.

  • Copied hinge modification to back door for easier access.

• Mini12864 V2.0 Screen Mount:

  • Replaced small stock screen with larger programmable LCD.

  • Designed 3D-printed mount under front door.

• Electronics DIN Rail Mount:

  • New motherboard didn’t fit stock plate.

  • Designed 3D-printed DIN rail mounts for flexible motherboard installation inside frame.

• Back Door Modification:

  • Built rear electronics bay door with hinges for quick access.

  • Integrated Noctua cooling fan and mini printed frame.

Result:

• Finished Voron build exceeded typical speeds, with improved flow rate and print reliability.

• Custom mods addressed airflow, thermal safety, usability, and serviceability.

• The printer now combines Voron’s ethos of DIY uniqueness with a set of tailored performance upgrades, making it both a high-speed and highly functional machine.

Computers and Software

AI Chatbot

Click the photo to find out more!

Problem:

I wanted to explore how to run a large language model locally without internet access, so the AI could process input and generate responses entirely offline. At first I didn’t know what I would use this project for, but it was a good way to practice Python and integrate different tools.

Method:

• Downloading the Model & Setup:

  • Chose DeepSeek as the LLM because it’s open source and easy to run locally.

  • Installed via Ollama.

  • Installed Python + pip, and added the libraries:

    • requests (HTTP requests)

    • pyttsx3 (text-to-speech)

    • PyAudio (audio input/output)

    • SpeechRecognition (voice-to-text)

    • json (data handling)

  • These allowed the AI to listen, understand, and speak back.

• Creating Code for the Model:

  • Wrote first Python script: chatbot loaded and responded to the text “hi.”

  • Basic interaction worked, but only one exchange at a time.

• Testing the Model:

  • Wrote second script: similar to first, but added text-to-speech output.

  • AI responded to “hi” and produced an audio file that played automatically.

• Talking to the Model:

  • Expanded script to add speech recognition.

  • This let me speak to the AI directly and hear it reply.

  • Difficult step, but enabled full voice interaction.

• Adding Wake Word & Kill Switch:

  • Added a wake word so the AI only listens when triggered (prevents it from responding to background sounds).

  • Added “end chat” command to return it to wake word listening mode.

  • Accidentally created a kill switch while testing, which turned into a useful feature to stop the program entirely.

Result:

• Completed a fully offline voice chatbot using DeepSeek + Ollama.

• System can:

  • Wake on command

  • Recognize speech input

  • Generate a response with the local LLM

  • Speak the response back with TTS

  • Return to listening mode or shut down on command

• A personal milestone project—my first major Python build—that proved I could integrate AI, audio, and interaction locally without relying on internet servers.

2012 Mac Pro Restoration

Click the photo to find out more!

Problem:

I inherited an old 2012 Mac Pro that was outdated and too slow for its original use. Instead of retiring it, I wanted to restore and modernize the machine—maxing out its hardware and turning it into a triple-boot workstation for tinkering with different operating systems, programming, and school work.

Method:

• Physical Modifications:

  • Cleaned and dusted the exterior and interior chassis.

  • Discovered it was already configured with dual CPUs, 8 RAM slots, and a GPU.

  • Reapplied thermal paste to refresh cooling.

  • Upgraded RAM by installing 8 × 16 GB DDR3 1333 MHz sticks → maxing out memory capacity.

• Software Modifications:

  • Installed two new 256 GB SSDs alongside the existing two 512 GB Apple SSDs.

  • Wiped all four drives for a clean setup.

  • Used OpenCore-Patcher to bypass Apple’s software block and install macOS Sequoia on one Apple SSD (primary boot).

  • Configured second Apple SSD as a backup.

  • Used Boot Camp Assistant from Sequoia to install Windows 11 directly (no USB needed).

  • Installed Linux onto one of the new SSDs using Balena Etcher.

  • End result: macOS + Windows + Linux all locally available, each with its own dedicated SSD.

    
    

Result:

• Successfully restored a 2012 Mac Pro into a triple-boot workstation.

• Hardware maxed out (128 GB RAM) and given new life with SSD storage.

• Runs macOS Sequoia, Windows 11, and Linux seamlessly.

• Machine is now reliable for programming, experimenting with multiple operating systems, and daily school work.