Suyog Khanal

02 · Transform

Experience

Multiple Internships and some Contract Work Experience

CareMagix — Software Engineer

Jun – Aug 2025

Remote

Led and mentored a team of 4 developers, driving delivery of website features and defect resolution.
Built a vector-embedded LanceDB database of 70K+ diagnostic (ICD) codes, queried to map sentences to relevant codes.
Programmed a custom Python class to access and query the database; tested and deployed it to production.

UBS — Data Engineering Intern

Jun – Aug 2024

Weehawken, NJ

Used Databricks to build data pipelines; ingested and transformed data with Python and SQL into clean tables.
Automated secret rotation for application accounts, integrating it into Azure pipelines for 100+ developers.
Designed and demoed a SharePoint site that routed forms into a shared Excel sheet via Power Automate.
Built a similarity-search tool to help employees find teammates; presented it to 30+ stakeholders.
Fixed a job-dependency bug from misconfigured synchronization, cutting potential delays by 50%.

Compunnel — Data Analysis Intern

Jun – Aug 2023

Plainsboro, NJ

Built a Customer Lifetime Value prediction model using machine learning algorithms.
Analyzed 500,000+ rows of data to calculate RFM scores, segment customers, and build Power BI reports.
Queried the LinkedIn API with Python for prospect research; used the OpenAI API to draft up to 5 emails/minute.

Commvault — Software Engineering Intern

May – Aug 2022

Tinton Falls, NJ

Managed virtual machines and used Python/PowerShell to run software and OS updates across them.
Wrote a test case to automatically update 50+ VMs and manage their snapshots.
Built a database table and Python backend to manage 100+ VM allocations, plus a reporting workflow.

03 · LOAD

Projects

Various projects made for personal and academic purposes

OBS Auto Launch

Aug 2026 – Present · github.com/SuyogKhanal5/obs-auto-launch

Built a cross-platform (Windows/Linux/macOS) desktop automation tool in Python that detects game launches across 5 platforms (Steam, Epic, GOG, Xbox/Game Pass, Battle.net) and drives OBS Studio's WebSocket API to automatically start/stop recordings.
Designed a platform-abstraction layer isolating ~600 lines of OS-specific logic (autostart, window management, process detection) behind a shared interface, letting one ~8,000-line core codebase run natively on all three operating systems.
Implemented self-healing OBS session management that detects a hung WebSocket connection or memory-bloated idle process and automatically kills/relaunches OBS with cooldown-based backoff to prevent restart loops.
Built a multi-track audio pipeline that clones the user's OBS profile to route mic/desktop/game/app audio to independent tracks, with per-application audio isolation that dynamically repoints capture sources and self-heals against an OBS quirk that silently rewrites them.
Wrote a shared Tkinter GUI installer (~1,100 lines) with OS-specific install flows — silent winget/Homebrew installs on Windows/macOS and a transparent preview-then-pkexec flow on Linux — plus safe update/uninstall detection that never overwrites existing settings.
Built a GitHub Actions release pipeline that cross-compiles native artifacts for all three OSes, including separate Apple Silicon/Intel runners to work around a numpy universal2 wheel limitation, with a rolling pre-release channel alongside tagged versions.
Wrote a 621-test suite (13 files) covering detection priority, audio routing, installer config generation, and clip trimming using a hand-built fake OBS WebSocket client, running the full suite in under a second with no real OBS/GUI dependencies.
Built an in-app video clip editor (VLC-backed, lossless and frame-accurate trim modes), a 40+ option settings GUI with live pickers, global custom keybinds for OBS actions, and a disk-space guard that inspects live audio meter data to catch failed recordings.
Grew and maintain a ~17,000-line Python codebase largely solo, including a phased migration from a Windows-only tool to a full Windows/Linux/macOS architecture.

PythonOBS WebSocketTkinterPyInstallerGitHub Actions

Rank Visualization ETL Pipeline

Mar 2026

Built an end-to-end player analytics platform pulling from the Riot Games API into custom graphs.
Automated ingestion with Snowflake stored procedures and Tasks that fetch and clean data on schedule.
Shipped a Streamlit-in-Snowflake dashboard with Altair visualizations for fully custom insights.

PythonSnowflakeSQLAPIsStreamlit

Esports Game Prediction Model

Nov – Dec 2024

Cleaned and analyzed ~100,000 rows of League of Legends esports data.
Built a model predicting game outcomes from only the first 15 minutes of data.
Wrote a 6-page paper detailing methodology and findings.

PythonSQLPandas

Street Fighter AI

Oct – Nov 2024

Trained an agent to play Street Fighter II using OpenAI Gym.
Implemented PPO with a custom reward function to shape fighting behavior.
Wrote a 5-page paper on the project and presented findings to 60+ people.

PythonOpenAI GymPPO

CNN Geoguessr

Apr – May 2025

Trained a CNN to predict a photo's country of origin from Google Street View imagery.
Live-demoed the model to 50+ people; wrote up findings in a 9-page research paper.

PythonGoogle Maps APICNN

Computer Vision Rhythm Game

Oct 2023 · HackRU Fall 2023

Built a rhythm game that generates a clap pattern and scores you on how well you remembered and reproduced it.
Used real-time hand tracking to detect claps from webcam video instead of a keyboard/controller input.

PythonOpenCVMediapipe

Tic Tac Toe

May 2023

Wrote a multithreaded server in C from scratch that manages a two-player online game of Tic-Tac-Toe.

CSocketsMultithreading

Shockwave

Jun 2022 – Present · addshockwave.com

Architect and maintain a ~38,000-line production Discord bot (Python, discord.py) spanning ~60 slash commands, with a 1,400+ test suite covering the full command surface.
Designed a self-healing SQLite schema that auto-migrates on every deploy — adding and backfilling columns as the schema evolves — with zero manual DB intervention across dozens of production schema changes.
Split game-record data (per-player, per-game Elo/win-loss) from account-wide economy data (currency, betting history) so a server can track multiple games' stats independently without cross-contaminating ladders.
Solved double-submission race conditions on Discord button interactions — e.g. two players clicking "Start" within milliseconds — using synchronous state flips before any await, preventing duplicate game starts without locks.
Implemented per-guild async betting-window timers that stay cancelable and reconcile cleanly across bot restarts, so an in-flight bet never gets silently orphaned.
Built persistent Discord UI views that survive restarts by reconstructing state from custom IDs and database lookups, guarded by a two-step confirm/cancel pattern on every destructive action.
Implemented context-aware autocomplete across 15+ command parameters and consolidated a sprawling command set into cohesive groups (/team, /tournament, /set, /make-teams) for discoverability.
Built a dynamic Pillow rendering pipeline that composites trading cards, brackets, and matchup graphics live from game state, plus a pari-mutuel betting economy and a full tournament system with seeded brackets, losers bracket, and Grand Finals for double elimination.
Built a hand-rolled Discord API mock harness to unit-test bot behavior without hitting the real API, running the full test suite in ~20 seconds via pytest-xdist.
Deployed and operate the bot in production on AWS Lightsail under systemd, with automated daily backups, structured logging, and a DM-based alert path for unhandled exceptions.

PythonSQLitediscord.pyPillowpytest-xdistAWS Lightsail

Education

Rutgers University – New Brunswick