Skip to content
View jaswin-codes's full-sized avatar

Highlights

  • Pro

Block or report jaswin-codes

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please donโ€™t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this userโ€™s behavior. Learn more about reporting abuse.

Report abuse
jaswin-codes/README.md

Hi, I'm Jaswin ๐Ÿ‘‹

Mechatronics @ UCT ยท UCT Racing EV Research lead ยท robotics, embedded & AI safety

Portfolio LinkedIn Email

I'm a first-year Electrical Engineering (Mechatronics) student at the University of Cape Town. I like building things that sit between hardware and software: robots, embedded systems and the AI that runs on them. I also do AI-safety research in short research sprints.

Open to vacation work and internships, Dec 2026 โ€“ Feb 2027.


๐Ÿ”ฌ Research

  • AfriGuard (Global South AI Safety Hackathon, Jun 2026). Safety evaluation of 4 frontier LLMs across 7 South African languages (1,120 adversarial prompts), with an auditable, language-aware scoring pipeline. Team of 4.
  • Beaten by a Cheap Surface Classifier (Apart Research, Digital Minds sprint, Aug 2026). Do LLMs predict their own behaviour better than an outside observer? A 21-feature surface classifier matched or beat the models' self-predictions in 6 of 10 conditions.
  • Evidence Before Erasure (Apart Research ร— CeSIA, AI incident response sprint, Sep 2026). Compared snapshot vs. event-driven evidence collection on a real trace of AI agents colluding on a wiki. Team of 5.
  • Hyperspectral freshness (Scientific Caribbean Foundation ร— UCT, May 2026). Simulation showing red-edge (760 nm) and NIR (1100 nm) bands predict produce freshness before visible spoilage. Now building a low-cost sensor for it.

๐Ÿ”ง Hardware & engineering

  • UCT Racing: leading the new EV Research subteam (electric drivetrains and battery chemistry), aiming for a Formula Student UK concept-class entry in 2027. Before that, power subsystem: cooling and telemetry research.
  • EMi Lab vacation work: MATLAB/Simulink/Simscape battery management system models (cell balancing, state of health) and a report on swappable batteries for electric tuk-tuks.
  • Water level alarm (EEE1009S): analogue design with LM324 comparators, CMOS logic and 555 timers, verified in LTspice and built on a breadboard.
  • Robotics: learning ROS 2 (first package) towards an autonomous mobile robot, and building an ESP32-C3 desk robot.

๐Ÿ’ป Software projects

  • jaswin-portfolio: my 3D portfolio site, built around an ESP32 model in Three.js, with a flat recruiter mode. Live at jaswin.me.
  • StatsGeeks: coordinate-aware random forest for Madrid โ†’ Amsterdam transfer learning. Won the Best F1 Score category at the StatsGeeks Data Science Hackathon.
  • jarvis-automation: voice-controlled desktop assistant with wake-word detection, voiceprint authentication and clap triggers.
  • jutsu_code_stage1: Naruto hand-sign combos recognised in real time with OpenCV and MediaPipe.

๐ŸŒฑ Currently

  • Learning ROS 2 and control systems
  • Co-founder and secretary of the UCT Maker Society; studio assistant at the d-school Makerspace
  • Treasurer, IEEE UCT Student Branch

๐Ÿ› ๏ธ Tools

Hardware & embedded
MATLAB Simulink LTspice Arduino ESP32 ROS 2 C

Software & ML
Python scikit-learn OpenCV TypeScript React Three.js Java

Pinned Loading

  1. StatsGeeks StatsGeeks Public

    Coordinate-aware random forest for Madrid โ†’ Amsterdam building-age transfer learning. Best F1 Score winner, StatsGeeks Data Science Hackathon 2026

    Jupyter Notebook

  2. jaswin-portfolio jaswin-portfolio Public

    My personal site, jaswin.me. A 3D ESP32 board you can click through, plus a quick-scan recruiter view. React, TypeScript and Three.js.

    TypeScript

  3. ubayd-hattas/AfriGuard ubayd-hattas/AfriGuard Public

    A multilingual AI safety benchmark evaluating frontier LLM safety across South African languages through automated red-teaming.

    Python

  4. UbaJaz/Digital_Minds_Research UbaJaz/Digital_Minds_Research Public

    Apart Digital Minds sprint (Aug 2026): a capability-controlled test of LLM privileged self-access

    Python

  5. ubayd-hattas/AI_Incident ubayd-hattas/AI_Incident Public

    Evidence Before Erasure: benchmarking evidence-collection policies for AI incidents (Apart X CESIA sprint)

    Python

  6. jarvis-automation jarvis-automation Public

    A voice-controlled desktop automation assistant with wake-word detection, voice authentication, and gesture triggers - built in Python. Version 1

    Python