Massah Arafeh

Experience

WHERE I HAVE BEEN

June 2026 — Present

SA Group

Software Engineer Intern

01Building a pharma ERP, owning the full stack solo: designed a 43-table PostgreSQL schema (procurement, traceability, inventory, finance) and a REST API(Next.js, TypeScript, 46 endpoints) with Prisma, Zod, and GitHub Actions CI; v1 delivered in 2 months.
02Made the compliance audit trail impossible to bypass with a branded TypeScript transaction type (an audit row written outside the mutating transaction fails to compile); added idempotency keys on financial and approval endpoints inside the same transaction, so a duplicate request under a race rolls back its own writes.

November 2024 — Present

University of Toronto Formula Racing Team

Driverless and Vehcile Dynamics Recruit

01Recovered 93% of teacher mAP50-95 (0.501 vs. 0.539, held-out validation) at 11.8× the throughput (381 vs. 32 FPS, FP32 TensorRT engines via ONNX) by distilling a YOLOv11x cone-detection teacher into a YOLOv11n student with feature-based knowledge distillation across the FPN neck, for NVIDIA Jetson deployment.
02Engineered a simulated LiDAR sensor in C# and built a ROS2 publisher node streaming point-cloud data across the driverless stack, following existing publisher architecture within a Docker + Unity environment.
03Built a PDF report pipeline using Jinja2 + Typst with sub-process error handling and lifecycle management, and a matplotlib visualization system producing 10+ plot types for vehicle dynamics study analysis.
04Designed KPI aggregation structures, extracting preprocessing configs, timestamps, and multi-variant outputs into unified data models, supporting dynamic metric selection by study type.

October 2025 — April 2026

BuildingAssets.ai

Software Engineer Intern

01Developed a multi-agent backend pipeline that processes machine label images and returns verified manufacturer documentation links in <30 seconds, eliminating hours of tedious manual web search for field auditors.
02Led OCR proof-of-concept using PaddleOCR; benchmarked accuracy and reported actionable recommendations
03Designed modular service architecture coordinating Gemini (OCR), Claude (reasoning), and Perplexity (search) accelerating processing speed by 1.6x

May 2025 - August 2025

Philer AI

Software Engineer Intern

01Led OCR pipeline development with Gemini 1.5, automating legal form extraction and cutting manual intake from 3–5 minutes to under 30 seconds, achieving F1 0.76 (Precision 79%, Recall 73%).
02Automated end-to-end data transfer from the codebase to Airtable via structured output integration and schema validation, eliminating manual handoff steps and reducing error-prone data entry across a team of 15+ engineers.

Selected Work

Follow along or reach out directly at
massah.arafeh@mail.utoronto.ca

ABOUT

ELSEWHERE