Adrit Panday.

Computer Science Engineer.

CS Specialist at the University of Toronto Scarborough with a minor in Astrophysics & Astronomy. I build at the intersection of machine learning, systems programming, and full-stack engineering — from neural networks in raw C to AI-assisted clinical imaging pipelines and real-time geospatial platforms. On the full-stack side I architect end-to-end products with Node.js, Next.js, and MongoDB, containerized with Docker and shipped with automated CI/CD. Outside of engineering, I'm usually deep in a video game or out in the park playing with friends.

View Work ↓

OCa~ PruRU:2#pB

02 / 04
Sunnybrook Hospital — Toronto, ON
Oct 2024 — May 2025

qc v~YWYOq! iRGmwCnwZ

Python · PyTorch · QuPath · OME-XML
  • Engineered automated preprocessing pipelines for histopathology tumor imaging, reducing manual data preparation time by 64% and enabling faster, more scalable model training workflows.
  • Built a multi-format image parsing framework supporting .svs and .tif whole-slide scans with QuPath and OME-XML integration, expanding dataset compatibility across research pipelines by 40%.
  • Optimized deep learning training pipelines with custom PyTorch classes for flow recomputation, geometric augmentation, and resolution normalization — improving training efficiency by 35%.
  • Enhanced AI-assisted tumor segmentation workflows, optimizing GPU-based preprocessing throughput for large-scale model training on high-resolution whole-slide images.
Directed inference pipeline: tile selection, neoplastic probability, cluster map
Inference Output
Annotated whole-slide histopathology image with labeled tissue regions
Annotated WSI
click to view

Z!y!Q5dq RSM2Q@j%

03 / 04
012025

kBF1

Full-Stack · Geospatial
Node.js · Next.js · MongoDB · Docker

Architected a full-stack geospatial platform enabling real-time visualization and analysis of Great Lakes nearshore water temperature data.

  • Built interactive Leaflet maps and Chart.js dashboards for real-time water temperature visualization across Great Lakes beaches.
  • Designed RESTful APIs with JWT-based authentication and MongoDB data models supporting high-volume geospatial data ingestion.
  • Containerized with Docker and implemented CI/CD via GitHub Actions with ~90% automated test coverage and production deployment readiness.
GLOW user dashboard where users can view and manage their saved beach locations and edit temperature logs
User Dashboard
GLOW Home Page with interactive map showing beach water temperatures across Toronto
Live Map
click to view
022026

RSG=C~ HIZ9Eo^ 79:Ppj

Systems · Machine Learning
C

Built a complete neural network training engine in C with no external ML libraries — implementing forward propagation, backpropagation, and explicit weight updates from first principles.

  • Derived and implemented full error backpropagation from scratch, carefully managing gradient flow and weight update sequencing across layers.
  • Achieved 80%+ classification accuracy on MNIST; analyzed performance gaps between shallow and multi-layer architectures on CIFAR-10.
  • Experimented with activation functions, hidden layer sizes, and input scaling to mitigate saturation and optimize convergence.
CIFAR-10 image classification samples across 10 categories
CIFAR-10 Samples
MNIST handwritten digit dataset samples used for training and evaluation
MNIST Training Data
click to view
032024

^J4 jQAxn +G7 V3 TMThlY@r

ML · Data Science
Python

1st place at MTA Datathon among 50+ participants. Applied ML-driven forecasting and budget optimization to build the highest-scoring F1 fantasy team within a $1M cap.

  • Analyzed 15+ race data points — track history, driver form, results, and car malfunction rates — to build performance forecast models.
  • Applied budget optimization within the $1M cap to identify the team scoring 193 points, securing 1st place among 50+ participants.
  • Collaborated in a 4-person team using agile workflows across data acquisition, model building, and final presentation.
F1 driver points and points-per-cost analysis bar chart
Driver Analysis
Optimal F1 fantasy team selection showing 193 total points within $1M budget
Optimal Team
click to view
042024

$$jc#gLF x w5N@Wtjl70Jh

Robotics · Computer Vision · AI
C/C++ · OpenCV · Linux · LEGO EV3

Built two autonomous robot systems from scratch — an adversarial soccer-playing robot and a probabilistic self-localizing navigator — both deployed on LEGO EV3 hardware under real-world noise conditions.

  • Developed autonomous soccer robot with webcam blob tracking, finite-state decision logic (ball intercept, goal pursuit, opponent avoidance), and Bluetooth actuation — placed Top 8 at RoboSoccer competition.
  • Built perspective-correction and heading-estimation calibration pipeline for a 170 cm × 115 cm playfield under noisy vision input.
  • Implemented Markov/histogram localization maintaining belief states across grid intersections × 4 orientations with probabilistic sensor fusion for robust recovery from ambiguous states.
Probabilistic localization grid showing robot belief states across orientations
Localization Grid
LEGO EV3 robot built for the RoboSoccer competition
EV3 Robot
click to view
052026

Ic;anqVo5G$ calORy^*Wg

AI · Algorithms
C

Engineered an adversarial multi-agent decision system combining A* pathfinding with MiniMax alpha-beta pruning for real-time grid navigation.

  • Built A* pathfinding with Manhattan heuristic and priority-queue optimization for optimal, real-time path planning in constrained grids.
  • Developed MiniMax with alpha-beta pruning, reducing evaluated states by ~60% at search depth ≥ 10 for sub-second adversarial decisions.
  • Designed the system to handle dynamic opponent behaviour, enabling robust navigation across large, unpredictable environments.
A* search order visualization showing explored nodes in a maze environment
A* Search Order
MiniMax adversarial search visualization showing agent and opponent search areas in a maze
MiniMax Search
click to view

mWFWC2U9Q b^HX1=

04 / 04
Languages
PythonCC++JavaScriptJavaSQL
Machine Learning & AI
PyTorchTensorFlowNumPypandasScikit-learnLLM APIs
Robotics & Vision
OpenCVSensor FusionLocalizationState MachinesMotion PlanningReal-Time Control
Frameworks
Node.jsExpressNext.jsReact
Tools & Infrastructure
GitGitHub ActionsDockerLinux (Ubuntu)MongoDBJiraVS Code
Familiar
HaskellAssemblyShellQuPathOME-XMLHTML/CSS
Computer Science Specialist
University of Toronto Scarborough
PythonPyTorchC / C++Next.jsDocker+ many more