Software Engineer @ Amazon · Toronto, ON

Mustafa Abdulrahman

I build GenAI agents, desktop apps, and the delivery pipelines that ship them safely to thousands of users.

Quest Log

Wait and hope.

S

Software Engineer @ Amazon

May 2026 – Present · Toronto, ONIn progress

Shipped an AI desktop assistant to 3,000+ finance analysts

  • Owned full continuous delivery for an Electron AI desktop assistant (Strands, Bedrock) rolled out to 3,000+ finance analysts — Playwright Windows E2E gates and daily prod canaries on EC2 (CDK), with zero manual approval steps.
  • Drove AppSec certification of a high-risk GenAI agent across 28+ security CRs, fixing 4 threat-model findings and 8 production risks via Bedrock Guardrails, per-user S3/IAM session isolation, a read-only SQL allowlist, and a sandboxed Python runtime.
  • Resolved Sev-2 incidents on call (silent Kinesis record loss) and root-caused a startup deadlock causing 0/18 successful launches by moving serial agent rehydration off a 90s readiness path into a worker pool.
  • Migrated Promise-to-Pay and Dispute workflows on a collections platform used by ~1,000 analysts from Redux-Saga to TanStack Query, and built group-level tasks for resellers with hundreds of linked accounts.
  • Ran user research with collections analysts to shape a consolidated multi-account inbox, and partnered with finance teams on agentic AI skills automating receipt matching and cross-country deductions.
  • Electron
  • Bedrock
  • Strands
  • Playwright
  • CDK
  • React
  • TanStack Query
A

Software Engineering Intern @ Amazon

May 2025 – Aug 2025 · Toronto, ONCompleted

Policy-to-code turnaround cut from 2 weeks → 2 days

  • Designed an end-to-end GenAI system (Bedrock, LangChain, MCP) that auto-converts approval policy PDFs into Java rules with test cases, reducing turnaround from 2 weeks to 2 days.
  • Implemented agentic workflows with TDD and LLMOps guardrails on Lambda, API Gateway, and CDK, eliminating 65% of potential production rollbacks and saving ~12 engineering hours per policy.
  • Bedrock
  • LangChain
  • MCP
  • Java
  • Lambda
  • CDK
A

Software Engineering Intern @ AMD

May 2024 – Apr 2025 · Markham, ONCompleted

GPU kernels in HIP 35% faster

  • Engineered a system using the GitHub GraphQL API to extract and store 2M+ entries in PostgreSQL with daily updates, visualized in Power BI reports shared weekly with upper management.
  • Optimized GPU kernel code in HIP — efficient matrix addition and moving-average filters — improving data processing speed by 35% and reducing compute time by 20%.
  • HIP
  • C++
  • GraphQL
  • PostgreSQL
  • Power BI
A

Machine Learning Engineer Intern @ MLDSAI Inc.

May 2023 – Aug 2023 · Toronto, ONCompleted

Manual task input cut by 50% with OpenAdapt

  • Developed the OpenAdapt Python library for AI-first process automation, using transformer completions for synthetic input generation to cut manual task input by 50%.
  • Scripted automated recording of user inputs and screenshots, generating a 10,000+ sample dataset and saving ~40% time over manual collection.
  • Accelerated data processing by implementing tokenization, reducing processing time by 30%.
  • Python
  • Transformers
  • OpenAdapt

Projects

Capstone

RabbitHole

Pick a seed paper and explore its citation neighbourhood as an interactive graph that ranks and explains how each paper relates.

  • Built the relevance engine: SPECTER2 embeddings of chunked full-text (max-pooled per paper) blended with year proximity and citation signals, with an embedding cache for fast re-runs.
  • Implemented bidirectional citation-graph expansion (references and citations) and the Flask API + Celery/Redis task queue that ingests papers asynchronously, with Flower monitoring in Docker Compose.
  • Designed the React Flow graph UI (dagre layout, relevance-coloured nodes, LLM explanations per edge) with live polling for background scoring jobs.
  • Benchmarked GPU (ROCm) vs CPU embedding throughput to size the ingestion workers.
  • Python
  • Flask
  • Celery
  • Redis
  • Neo4j
  • SPECTER2
  • React
  • React Flow
  • Docker

GitHub ↗

Hack the North 2025 · Semifinalist

CaptureCube

A motorized photo cube that shoots a product from every angle, then turns those shots into an AI-generated ad video.

  • Built the web app end to end: a React + TypeScript studio with an interactive three.js 3D product preview, video preview, and one-click download.
  • Wrote the Flask backend that pulls the Raspberry Pi's captures from S3 and drives Google Veo 2 (Vertex AI) image-to-video generation, polling long-running jobs to completion.
  • React
  • TypeScript
  • three.js
  • Flask
  • AWS S3
  • Vertex AI · Veo 2
  • Raspberry Pi

GitHub ↗Devpost ↗

Hardware Sign-out Site

Open-source inventory & applications platform for Canada's largest Makeathon.

  • Directed a 15+ person team building an open-source inventory system for 5,000+ components — hardware wait times went from 4 hours to 10 minutes.
  • Built an application review portal that cut screening time from 20 hours to 2 hours for 1,500+ applicants.
  • React
  • Redux
  • Django
  • PostgreSQL
  • Docker

GitHub ↗

OpenAdapt

Open-source library for AI-first process automation using large multimodal models.

  • Core contributor during my MLDSAI internship: synthetic input generation with transformer completions and automated input/screenshot recording.
  • Python
  • LLMs
  • Automation

GitHub ↗

Trophy Room

Hackathons, won and fought.

3 hackathon wins · all projects on Devpost ↗

  • NCAR

    Winner · Transport & Travel · MakeUofT 2022

    Nighttime Collision Avoidance System: an affordable driving assistant that spots people on a pitch-black road with an infrared camera and a TensorFlow model, then warns the driver by projecting onto the windshield.

    • Raspberry Pi
    • TensorFlow
    • OpenCV
    • Infrared camera

    Devpost ↗GitHub ↗

  • FocusHacks

    Winner · Most Creative Use of Twilio · NewHacks 2021

    An ML study buddy that watches for when you lose focus and emails you a nudge to get back to work.

    • TensorFlow
    • Node.js
    • Express
    • Twilio SendGrid

    Devpost ↗GitHub ↗

  • Team Jungle

    Winner · MLH Fellowship Orientation Hackathon 2022

    A modular team portfolio site driven entirely by a JSON file, with an interactive map of everywhere the team has been.

    • Flask
    • Jinja
    • Folium
    • Bootstrap

    Devpost ↗GitHub ↗

  • CaptureCube

    Semifinalist · Hack the North 2025

    Low-cost product ads in minutes: a dual-axis photo rig plus AI video generation.

    • React
    • Flask
    • Raspberry Pi
    • Veo

    Devpost ↗GitHub ↗

Spellbook

Technical skills, ranked by the seven tiers of magic.

Languages

  • TypeScript / JavaScriptSaint
  • PythonSaint
  • JavaAdvanced
  • C++Advanced
  • SQLAdvanced
  • HTML / CSSSaint

Frameworks

  • ReactSaint
  • TanStack QueryAdvanced
  • ReduxAdvanced
  • ElectronAdvanced
  • PlaywrightAdvanced
  • DjangoAdvanced
  • Spring BootIntermediate
  • FlaskIntermediate

AI & Agents

  • Amazon BedrockSaint
  • Strands AgentsAdvanced
  • Bedrock GuardrailsAdvanced
  • LangChainAdvanced
  • MCPAdvanced

Cloud & Tools

  • AWS (S3, IAM, Lambda, EC2, CloudWatch)Saint
  • AWS CDKAdvanced
  • DockerAdvanced
  • CI/CD · JenkinsAdvanced
  • PostgreSQL / MySQLAdvanced
  • GitSaint

Education

University of Toronto

Bachelor of Applied Science in Computer Engineering

Graduated May 2026