Mithun
Sivapathasundram
Acomputersciencegraduatewithapassionforbuildingclean,performantsoftware.
Interestedinfull-stackdevelopment,machine learning,andtheintersectionofgreatengineeringandgreatdesign.
Based in
Toronto, ON
Education
BSc Computer Science @
Toronto Metropolitan University
Graduated · 2026
Looking for
2026 New Grad Roles
Software Engineer Intern
May 2025 – Dec 2025Environment and Climate Change Canada · Toronto, ON
Built features and hardened data integrity on the Environmental Emergency Regulations reporting system, where 12,000+ facilities across Canada report hazardous substance holdings.
- Built 15+ features on a 3-developer team for the Environmental Emergency Regulations reporting system, a C#/.NET 8 platform where 12,000+ facilities across Canada report hazardous substance holdings: facility transfer workflows, bulk import tooling, automated notifications
- Wrote the MS SQL stored procedures that detect and repair corrupted facility records, fixing 4,000+ records and retiring a weekly manual validation check
- Added .NET validation rules and MS SQL constraints across the reporting intake path so malformed submissions are rejected at entry rather than downstream
- Diagnosed and fixed production defects across the frontend and .NET services, raising Azure DevOps-reported test coverage from 65% to 85%
Software Engineer Intern
Jan 2024 – Sep 2024Ontario Treasury Board Secretariat · Toronto, ON
Shipped full-stack features and data automation for the legacy web applications that guide Ontario Ministry of Finance staff through tax appeal workflows.
- Shipped full-stack features across two legacy web applications that guide Ontario Ministry of Finance staff through tax appeal workflows: required steps, deadlines, and the parties to notify at each stage
- Added client-side validation and REST API integrations to those applications, standardizing how appeal data was entered and cutting downstream processing errors
- Designed the relational schema and query layer behind an automated ETL pipeline that replaced a weekly manual appeals report with same-day delivery to Ministry of Finance analysts
- Built a library of reusable, responsive front-end components in custom CSS from Figma specs, working with UI/UX designers from wireframe to production
- Built and documented the REST APIs behind the appeals applications, gated behind automated tests in CI/CD before merge
- Built a classifier on Power Automate's AI Builder that flagged AI-written job applications, placing 2nd of 30+ teams at an internal hackathon
Fantasy Basketball Companion
An LLM coaching app that recommends fantasy basketball roster moves, built solo end to end.
- →Designed and built the full stack solo in Next.js and TypeScript: an LLM coaching app that recommends fantasy basketball roster moves
- →Split the original Next.js/Supabase monolith into three independently deployable services on GKE Autopilot using Docker multi-stage builds, with a HorizontalPodAutoscaler scaling the AI inference service 1 to 5 replicas on CPU load so inference cost scales separately from the stateless frontend
- →Backed the app with REST endpoints over JSON, a PostgreSQL data model, and Redis caching for repeated queries, with tests gated by GitHub Actions
- →Wired LangChain.js with ConversationBufferMemory so the coaching engine keeps context across roster questions, streamed to the client
NBA Predictive Analytics Platform
End-to-end ML pipeline predicting per-game stat lines for 500+ active NBA players.
- →Built an end-to-end ML pipeline that predicts per-game stat lines for 500+ active NBA players, combining six ensemble models with automated daily retraining on fresh season data
- →Built a leakage-free walk-forward eval harness over 3 backfilled seasons (79k player-games); trained it on real outcomes to beat the baseline by 2.9% MAE (4.59 vs 4.73, p < 0.001) across three validation protocols
- →Built a Python ETL job on GitHub Actions that scrapes and normalizes Basketball Reference statistics daily into a clean, versioned dataset, no manual steps
SimpleBL
Retrieval-augmented research assistant that grounds every answer in live PubMed literature.
- →Built and deployed a retrieval-augmented research assistant that pulls matching PubMed articles per query and grounds every answer in those sources via LangChain and the Groq API
- →Wrote a 340ms inter-request limiter to stay under NCBI's 3 req/s policy, with graceful degradation so PubMed outages surface as a clear message rather than an unhandled error
- →Rendered each answer beside its source articles in a typed React client with centralized state, so users can trace any claim to a citation
VetConnect
School Project: Full-stack veterinary management system connecting pet owners and clinics.
VetConnect
- →Collaborated with 4 developers to design and ship a secure, real-time patient management platform handling appointments, records, and messaging.
- →Implemented role-based access control via Supabase Row-Level Security, ensuring strict data privacy between pet owners and veterinary staff.
Smart Recruiter
AI-powered recruitment platform. Placed 2nd at the OPS Phenomenal Hackathon.
Smart Recruiter
- →Reduced manual screening time by integrating Microsoft's Category Classification Model to flag AI-generated job applications.
- →Delivered a full end-to-end platform within a tight hackathon timeline using Microsoft Power Apps and REST APIs.
Olympic Medal Predictor
Linear regression model forecasting Olympic medal counts from 100+ years of historical data.
Olympic Medal Predictor
- →Trained and evaluated the model across multiple Olympic Games, achieving strong predictive accuracy for country-level medal outcomes.
- →Built end-to-end data cleaning, feature engineering, and interactive visualisations using pandas, seaborn, and scikit-learn.
Soccer League Database System
School Project: Relational database system managing all data for a competitive soccer league.
Soccer League Database System
- →Designed a normalised Oracle DB schema covering teams, players, fixtures, results, and standings with full referential integrity.
- →Conducted end-to-end testing to verify data consistency and reliability across concurrent read/write operations.
Java Stock Market System
School Project: OOP stock market simulator with real-time data, built by a team of 6.
Java Stock Market System
- →Engineered comprehensive UML models to drive a clean, extensible OOP architecture across a 6-person team.
- →Integrated the Alpha Vantage API for live stock data and achieved full unit test coverage with JUnit.
Languages
Frameworks/Libraries
Databases & Cloud
Tools & Workflows
Testing