Aarib Sami
CS @ University of Guelph • full-stack • AI + SaaS
about
I like building, learning AI and shipping SaaS products.
Recently I’ve especially been focused on agentic AI especially with all the new tools coming out like OpenClaw.
experience
Software Developer Intern
Candescent • Waterloo, ON
May 2025 – Dec 2025
- Optimized file processing algorithms in Spring Boot, reducing average API response times by 15% and eliminating timeout errors for 500+ concurrent users during peak upload periods
- Built a bulk wire payment system using React.js and TypeScript, simplifying batch payment processing for 100+ companies through automated text file upload and customizable field mapping to database schemas
- Defined server storage allocation and upload limits by benchmarking CSV parsing performance across file sizes and formats, reducing storage overhead while preventing timeout failures for major financial institution clients
- Found critical edge cases in field mapping logic pre-release by writing Jest unit tests, achieving 90% test coverage
projects (click to expand)
furnish.ai React (TypeScript), Node.js, Supabase, Stripe
- Built a full-stack AI SaaS using React, TypeScript, Node.js, and Express that digitally furnishes empty room photos for real estate photographers in 30–60 seconds, onboarding 15+ users within the first week of launch
- Integrated Stripe Checkout API across 5 pay-as-you-go pricing tiers ($8–$300) with automated webhook handlers to instantly credit user generation balances upon purchase
- Distributed organic content through OpenClaw AI agents generating/posting before-after staging creatives using GenViral API, resulting in 300 Instagram followers with $0 paid promotion
Policy Compass Python, FastAPI, React, TypeScript, Qdrant, Gemini, Inngest
- Created a searchable company policy knowledge base by building a RAG pipeline that chunks PDFs into 1,000 token segments, storing 3,072-dimension Gemini embeddings in Qdrant, powering quick sourced LLM answers
- Enabled self-service policy lookup through a FastAPI REST layer with throttles (20 queries/min, 5 ingests/hr) and a React + TypeScript frontend, resulting in safer rollout and controlled API spend
- Achieved zero failed responses during Gemini rate limiting and outages by implementing durable Inngest workflows with retries at each step and a fallback chain across 3 models
- Eliminated duplicate vectors on re-ingestion by batching up to 100 texts per Gemini API call and assigning UUID5 deterministic chunk IDs for idempotent Qdrant upserts, keeping the vector store clean across policy updates
FireGuard AI Python, PyTorch, Flask, React.js, HTML/CSS
- Trained a two-layer feedforward neural network using PyTorch by preprocessing 6,000+ meteorological records and engineering 5 weather features for forest fire classification, achieving 88% accuracy on test data
- Built a React frontend using react-map-gl and Mapbox satellite imagery enabling users to click anywhere on the globe to assess forest fire risk, providing actionable risk classifications and confidence scores
- Designed a Flask REST API with OpenWeatherMap integration that ingests live weather data, extracts meteorological features and executes model inference, returning tiered risk classifications and explanations
GryphCourseWatch Python, Flask, Playwright, HTML/CSS, JavaScript
- Built a system for students to monitor university course availability using Python, Playwright, and Flask, resulting in a 30% reduction in manual registration time across 100+ tracked sections
- Built a backend notification flow using Flask APIs and SMTP email delivery, resulting in alerts reaching students within 5 minutes of seat availability detections
- Led a pre-launch campaign on LinkedIn, generating interest from 1,500+ students, validating campus demand
very old projects
FaceID Attendance System Python, OpenCV, Firebase
- Designed an attendance system using Python and OpenCV, achieving a facial recognition accuracy of 95%
- Deployed automated attendance logging by detecting faces within 1.75s and updating the database accordingly
- Leveraged Firebase to dynamically update student attendance data, ensuring accurate records for future use
- Built a frontend interface using OpenCV's graphical system to display student data such as name and major
Sign Language Translator Python, OpenCV, TensorFlow
- Built a live sign language recognition system using TensorFlow, achieving 92% accuracy across 26 gestures
- Utilized OpenCV for efficient video capture and processing, enabling detection within ~1.53s latency/frame
- Fine-tuned an SSD MobileNet model, boosting recognition performance by over 25%
Contact List Organizer Python, Flask, PostgreSQL, HTML/CSS
- Designed and developed a contact management system to facilitate CRUD operations on contact data
- Utilized HTTP methods to perform corresponding CRUD actions on contact resources
- Implemented email functionality to send bulk emails to selected contacts directly from the app, decreasing time required to send bulk emails by 23%
- Used SQLAlchemy ORM for efficient data management with a PostgreSQL database
Zoo Animal Classifier (ML) Python, Machine Learning
- Built with a zoo dataset of 100 animals, each with 10 attributes and a class label
- Implemented a k-nearest neighbors (k-NN) model to predict class labels for new animals
- Used Euclidean distance, Hamming distance, and Jaccard similarity for evaluation
- Achieved 70% accuracy on a 20-animal test set
CarBase C
- Developed a dynamically linked list implementation in C for car data management
- Implemented core operations including adding, editing, and removing car information
- Used user-defined structures and pointers to enable efficient record manipulation and traversal
- Demonstrated strong memory management and data-structure design
Alpine Escape Unity, C#
- Built an endless-runner mobile game in Unity and C#
- Designed game visuals including character, obstacles, and environment assets
- Implemented core mechanics such as obstacle avoidance, scoring, and progression
Humanitarian Aid Service Java, Swing
- Collaborated on a Java-based aid request platform simulation
- Built Swing interfaces for reviewing and managing loan requests
- Used object serialization and application-level data modeling for persistence
- Implemented filtering, status updates, and dynamic list rendering
Lazer Defender Unity, C#
- A 2D mobile game developed in Unity using C#
- Player controls a spaceship and avoids/eliminates enemies
- Goal is to maximize score
Glitch Garden Unity, C#
- A 2D Plants vs Zombies-inspired game built in Unity with C#
- Place defenders strategically to stop enemies that increase in speed over time
- Resource management mechanics tied to defender placement costs
Catch The Icicle Unity, C#
- A 2D mobile game developed in Unity using C#
- Catch falling icicles to increase score
- Primary objective is score maximization