Hi, I’m Rashid.

• studying

  Third year of ECE (PEY Co-op) at U of T. This term: Operating Systems, Data Structures & Algorithms, and Control Systems.

• just wrapped

  A summer at LGND AI doing geospatial machine learning on satellite-image embeddings.

• building

  A searchable, source-linked record of human-rights incidents, extracted from 40,000+ scanned newspaper pages.

1. May – Aug 2026

   Engineering Intern, Geospatial ML at LGND AI

   Remote

   • Shipped 10+ projects on a two-week cadence with no pre-assigned tasks, mapping what LGND’s satellite-embedding platform could and couldn’t do: national-scale analyses, few-shot detectors, four FastAPI/Leaflet web apps, and two benchmarks.
   • Reached 65% of Microsoft’s published U-Net F1 on poultry-facility detection with a few-shot linear probe. It trained in 2 minutes on 80 labelled chips, compared with the U-Net’s 6,013 labels and 42 TB of imagery.
   • Cut a continental land-cover map’s storage by 98% (24.6 GB → 456 MB) with PCA-64 and int8 quantization, after showing the loss was smaller than the existing label noise.
   • Built a national similarity-search app that recolours all 55,793 U.S. cells on every click with zero API calls, by shipping every embedding to the browser as a single 14.7 MB payload.
   • Found that random k-fold cross-validation overstated detector precision by 23 points compared with leave-tile-out, so I moved all evaluation to spatial hold-outs and re-measured every headline number.
2. Jun 2026 – now

   Volunteer Software Engineer at a national non-profit

   IT Department · Toronto

   • Built and launched a Spring Boot + SQL app that 200+ members of a 30,000+ member national organization use to propose, approve, and schedule every event.
   • Implemented the review flow (one submission form, an admin queue to accept, reject, or propose a new date, a shared calendar, and automatic email confirmations), plus authentication and role-based access control.
3. Sep 2024 – Jan 2025

   Embedded Software Developer at UTRA Combat Robotics

   University of Toronto Robotics Association

   • Wrote C++ firmware for the team’s combat robots, covering sensor integration and motor control, and fixed control-loop failures found in testing to stabilize motor response.
   • Reviewed teammates’ firmware to keep modules separate and the codebase maintainable as it grew.

2026Bare-metal CNios V RISC-VDE1-SoC FPGA

Real-Time Fractal Visualizer

Renders Mandelbrot and Sierpiński fractals to VGA in real time, in bare-metal C on a Nios V RISC-V soft-core. Floating point was too slow for real-time frames, so I replaced it with hand-written Q8.24 fixed-point math, then added 4×4 block rendering, zoom-adaptive iteration counts, and double-buffered vsync.

I wrote the PS/2 mouse driver from scratch for click-to-zoom, and added an int8-quantized CNN gesture recognizer for touchless zoom and pan.

With Tommy Yu for ECE243 · course code kept private

Watch the demo ↗

2026C++EZGL/GTKAlgorithms

OpenStreetMap Mapping & Routing

Parses city-scale OpenStreetMap data into custom C++ data structures, with containers and indexes chosen so every query (name search, street length, one-way connectivity) meets strict complexity limits. Routing uses Dijkstra and an optimized A*, which let our team pass every routing performance test.

A multithreaded solver for an NP-hard travelling-courier problem builds a greedy route for ~5 s, then spends the remaining ~45 s improving it with simulated annealing.

Team of three for ECE297 · course code kept private

Watch the demo ↗

2026PythonOpenCVGemini/OpenAI

Human Rights Incident Database & Map

A searchable platform (interactive map, categorized views, and a state-policy tracker) documenting persecution incidents since 1947. The records were extracted from 40,000+ scanned newspaper pages, 35,000+ of them in Urdu. Every incident links to its source page and was hand-checked.

The layouts varied too much for templates, so an OpenCV pipeline splits each page into articles by detecting its rule lines. That turned out to matter: on 1950s Nastaliq newsprint, whole-page OCR invented words, while clean per-article crops came back accurate. A recall-tuned Gemini/OpenAI stage then finds and translates incidents. Its context is capped at two pages to curb hallucination, and per-page checkpoints let failed runs resume where they stopped.

Ongoing personal project

2026PythonFastAPILeafletLGND API

Reverse GeoGuessr

A geography game built on LGND’s embedding API before my internship started. You read a description of a place (“a wildfire burn scar next to still-green forest”), click the map, and get scored on how closely your spot matches in embedding space.

Free server, so it can take ~40 s to wake up

Watch the demo ↗ Play it ↗

More on GitHub ↗

Languages

C, C++, Verilog, Python, Java, SQL, JavaScript

Embedded

Bare-metal C on Nios V RISC-V, DE1-SoC (Intel Cyclone V), PS/2 and VGA drivers, fixed-point math, Arduino

Tools

Git, Linux, Docker, GDB, Valgrind, ModelSim, Quartus, Snowflake

Data & ML

NumPy, pandas, FastAPI, OpenCV, embeddings and vector search, PCA and int8 quantization, spatial cross-validation

Bench

Oscilloscopes, waveform generators, multimeters, DC power supplies