Available for work

Hi, I'm Arnob Ghosh
Software Developer.

I build data-visualization tools, computer-vision pipelines, and API integrations that ship. BSc (Honours), Computer Science — Memorial University of Newfoundland. Equal parts curiosity about emerging tech and discipline about making it actually useful.

Arnob Ghosh portrait
// featured

Proper Noun Explorer

// skills

Summary of Qualifications

Angular Pandas NumPy Bootstrap

With a strong foundation in modern web development, I specialize in building dynamic, responsive applications that deliver seamless user experiences. My expertise includes crafting interactive front-end solutions and leveraging powerful data-analysis techniques to extract meaningful insights. I thrive in developing computer-vision applications and implementing machine-learning workflows to solve real-world problems. Whether designing sleek, mobile-friendly interfaces or processing complex datasets, I bring a versatile skill set that bridges the gap between data science and practical software development.


// education

Education


// experience

Work Experience

Independently scoped, built, and shipped — engagements where I owned the design and delivery end-to-end. Click any thumbnail to enlarge.

Dec 2025 — Present

Full-Stack Engineer · Rosterly

Document intelligence & calendar automation
  • Built a web application that converts staff-roster images into structured, editable tables via OCR and pushes approved schedules into Google Calendar.
  • Designed a human-in-the-loop review surface that flags low-confidence cells for correction before downstream sync — prioritising correctness and auditability over raw OCR accuracy.
  • Implemented an idempotent sync layer against the Google Calendar API, so the same approved roster can be replayed without producing duplicate events.
  • Owned the design and delivery from OCR extraction through to the user-facing review UI and calendar integration.
Python OCR Google Calendar API Web frontend Workflow automation Human-in-the-loop
Rosterly end-to-end demo
Sep 2025 — Dec 2025

Identity & Integration Engineer · Conduit Systems

Backend integration tooling & SaaS administration automation
  • Engineered a FastAPI-based proxy service that intercepts and rewrites outbound HTTP traffic, restoring egress from sandboxed runtimes by injecting compliant request headers and defeating edge-layer rejections.
  • Designed and implemented an administration framework over Atlassian's organization-level APIs to automate end-to-end user onboarding, group assignment, and deprovisioning across Confluence workspaces.
  • Reconciled dual authentication models (organisation tokens and admin tokens) and mapped human-readable group names to internal Atlassian identifiers.
  • Exposed the proxy securely via ngrok with persistent session continuity for upstream consumers.
Python FastAPI REST APIs Atlassian ngrok Google Apps Script IAM
Conduit Systems architecture overview
May 2025 — Sep 2025

Computer Vision Engineer · Vergence Vision

Camera calibration & pose-estimation tooling
  • Built an end-to-end computer-vision toolkit in Python that elucidates camera calibration and pose estimation using OpenCV ChArUco markers.
  • Authored auxiliary image-processing utilities for autonomous cropping, resizing, undistortion, and color-space conversion (raw → color → grayscale → binary).
  • Documented version-pinned dependencies (OpenCV Contrib 4.8.1) and published a project page so external collaborators can reproduce results.
Python OpenCV Contrib NumPy Image processing Camera calibration Pose estimation
ChArUco marker detection

// side projects

Projects

Smaller experiments and places I went to learn an idea, a library, or a language. My interest in computer science is in data visualization; if physics can be bifurcated into classical and modern, my interest is in the classical. The projects below, done in groups or individually, reflect those interests.

For any group project mentioned here, I do not expect anyone to be credulous enough to believe my contribution from my words alone. The commit history of each repository should corroborate the claims I make. If a description seems equivocal, please contact me for clarification — my contact details are at the bottom of this page.

Check out my biggest Python project so far, where I tried to implement almost all of my concepts about the tree data structure. It has exponential runtime but ensures optimal solutions — unlike genetic algorithms, for which I prefer Java. A few smaller projects I made while using pandas:

  • gistemp — visualised NASA data with matplotlib (pyplot, colors).
  • Cary — visualised three datasets about Cary, North Carolina in one notebook. (Why isn't the Cary Arts Centre counted as a point of interest?)
  • pca — attempted to recreate Figures 3 and 5 from a breast-cancer classification paper; flagged axis-labelling issues in the original.
Tic-tac-toe game screenshot

Brython Tic-Tac-Toe

Tabannum and I developed a tic-tac-toe game using the Brython library of Python. Available upon request — feel free to contact either of us.

Search Algorithms (JS)

I have written search algorithms — BFS & DFS, A*, IDDFS — without any use of recursion, in JavaScript. They are the rough algorithms only; the environment, configuration, etc. are not included.

FishFrenz game screenshot

FishFrenz (ProjectAI)

A snowy night. The crocodile is trying to eat your fish friends, who take erratic turns every frame. The voracious enemy targets one fish — its direction of motion catalysed by the fish it stalks. Built in JavaScript with Zawad Mutawassit.

Project Catalina430 — Geometry Wars-style game

Project Catalina430

Catalina430 created a game analogous to Geometry Wars in SFML — the same library I used for my first C++ project. I added GUI elements and made changes to the SFML core. Rather than fork-and-contribute, I named it a disparate project.

Bouncing shapes C++ project

Bouncing Shapes

My first C++ project using SFML and ImGui, inspired by a YouTube tutorial.

JavaSwing1

A Java Swing project that lets you fill a 9×9 grid with tiles of different shapes and colors. The biggest challenge was ensuring tiles cannot overlap — once that invariant held, the rest of the interaction (choose a shape, preview placement on hover, click to place) became straightforward.


// playable demo

Mega Man ✕ Mario, in your browser

A small canvas platformer experiment. Click to load; use the controls below.

Your browser does not support the HTML canvas tag.
W — Jump A — Run left D — Run right


// contact

Let's build something.

The fastest way to reach me is email. I respond within a day.