Yash Kathrani

Challenge

VEX's 2024-25 "High Stakes" season: score rings on stakes (up to 3 points each) and climb a central ladder structure (up to 12 points) on a 12x12ft field, inside an 18x18x18in robot size limit.

Design

As team captain of Team 19359A, I led the mechanical design and directed the robot's programming, working with our team to brainstorm scoring mechanisms including an over-under flex-wheel intake, hook- and hood-style mobile-goal scoring, and a first-class-lever clamp to grab mobile goals from the field edge, targeting roughly 450 RPM through a 36-48 tooth drivetrain ratio for a ~10in wide base.

CAD

Modeled the full robot in Fusion 360 subsystem by subsystem: drive base, clamp, wall risers, a two-stage intake, a wall-stake scoring mechanism, and a climbing mechanism, each iterated through multiple CAD revisions (the clamp alone went through a V1 and V2) before being built.

Build

Manufactured and assembled each subsystem in sequence, including a ratchet mechanism for the climb, integrating them into a complete competition robot.

Controls

I led development of the control software: programmed in C++ on the PROS framework with the LemLib motion library, tuning PID control for autonomous movement and driver-assist macros.

Testing & Iteration

Built and refined multiple autonomous routes (a solo autonomous win-point route, positive- and negative-side routes, a goal-rush route, elimination routes, and a dedicated skills route), developed a custom route-visualization tool to plan paths, and logged technical difficulties and fixes after every competition.

Competition

Competed through a full season of qualifiers and skills events and qualified for the VEX Robotics World Championship.

Tech Stack

Engineering Notebook ↗

Challenge

FTC's "Into the Deep" season: intake game samples from a submersible play area and score them quickly and precisely, with cycle time as the main constraint.

Design

As team captain of Cyber Tech #20720, I led the mechanical design and directed the robot's programming, designing a mecanum-drive robot around two main subassemblies with our team: a sample intake and a linear-slide scoring arm, built to work around the game's central obstacle, the submersible.

CAD

Modeled the full robot digitally, then validated the design with laser-cut wood chassis prototypes to confirm the CAD model was accurate before committing to final materials.

Build

Custom-CNC'd the mecanum chassis out of aluminum for stability, with pocketing to cut weight while preserving structural integrity for the linear slides. Built a two-hook climbing mechanism on vertical linear slides, driven by 435 RPM motors.

Mechanism Design

Iterated the intake through surgical-tubing, boot-wheel, and Gekko-wheel designs before settling on 72mm Gekko wheels (driven by 290 RPM servos) feeding a scoring arm mounted on a 4-stage Misumi linear-slide system, extended by a servo-powered 5-stage scissor linkage.

Controls

I directed the control software: programmed autonomous and driver-assist behavior in Java using Roadrunner and PID heading control for drift-free movement, with TensorFlow Lite and OpenCV for object detection and REV through-bore encoders for accurate linear-slide positioning.

Testing & Iteration

Used MeepMeep to visualize and tune trajectories, adjusted PID constants (kV, kA, kStatic) to match measured against target velocity curves, and iterated based on competition results, including switching to custom-trained object detection and upgrading to more accurate through-bore encoders after early accuracy issues.

Competition

Consistently scored 40+ points in the 30-second autonomous period and won Tournament Champion as Winning Alliance Captain, qualifying for the state tournament.

Tech Stack

Engineering Notebook ↗

Overview

Brailed is a phone-mounted Braille keyboard and agentic smartphone controller our team built to make smartphone interaction more accessible and more private.

Problem

Touchscreen interaction is a real barrier for blind and low-vision users, and voice control isn't always private or practical in public settings.

System Architecture

Hardware

Six capacitive touch sensors represent Braille inputs, read by an ESP32 that handles chord detection, with an OLED providing device status and feedback, all built into a custom 3D-printed frame that mounts to the phone.

Software

The ESP32 communicates with the phone over BLE GATT, and an Android companion app (Kotlin) translates Braille chords into phone actions. AI-driven commands run through the Jac language and byLLM, constrained by structured capability restrictions so model output can't directly trigger unrestricted phone actions.

Engineering Challenges

Reliable capacitive chord detection, keeping BLE communication responsive enough for real-time typing, and constraining AI output to a safe, structured action set were the core problems our team solved.

Result

Won 2nd Place at JacHacks SF 2026.

Tech Stack

Overview

CHEHRA is a physical facial-recognition attendance system our team built and deployed for real classroom use, developed across 2025 and 2026.

Problem

School check-ins are slow and outdated: students forget IDs, lines back up, and staff lose time manually tracking who's present.

Hardware

An Orange Pi drives a camera and display, mounted in a custom Fusion 360 enclosure with dedicated camera and display mounts. The hardware went through more than ten enclosure iterations.

Software

Python and OpenCV handle facial recognition on the Orange Pi. A React Native app provides the interface, and Supabase stores registered users and attendance logs.

Engineering Challenges

Camera angle, enclosure fit, classroom lighting, component positioning, startup reliability, logging reliability, and recognition accuracy all had to be solved for a device that runs unattended, every day, in a real classroom.

Results

Approximately 96% recognition accuracy, more than 250 classroom check-ins logged, and testing through two real classroom pilots.

Tech Stack

View on Devpost →

Overview

PillPass is a facial-recognition medication dispenser our team built in about 36 hours at CruzHacks 2025.

Problem

Seniors managing multiple medications need a dispenser that recognizes the right user and enforces the right schedule, without added complexity.

Hardware

A Raspberry Pi handles facial recognition, while an ESP32 and Arduino coordinate a stepper-motor-driven dispensing mechanism and a TFT display.

Mechanical Design

The chassis was designed around the pill channel, the stepper motor, the display, and the electronics, with serviceability in mind. The pill pathway was kept physically separate from the electronics bay so nothing jammed or interfered, and the motor and display mounts were iterated quickly enough to assemble, test, and adjust during the hackathon.

Engineering Challenges

Keeping the medication pathway aligned with the motor-driven mechanism while leaving enough room for the electronics was the central mechanical challenge, on top of the usual hackathon time pressure.

Result

Won Best Hardware Hack at CruzHacks 2025.

Tech Stack

View on Devpost →