Hi! / Bonjour! / Hola! / Привіт! 👋
AI Engineer, Researcher, and Entrepreneur. In my free time I enjoy football, lifting weights and muay thai.
AS

About

Hi, I’m Axel Sanchez, an AI Engineer at Scotiabank on Scotia Intelligence, the bank’s central AI team, and a Software Engineering graduate of the University of Calgary. I specialize in high-performance systems ranging from production AI agents and GPU computing to cloud microservices and blockchain applications. Before Scotiabank, I worked at Microsoft, Aptum and Sigma. Take a look around, and if something resonates, I’d love to connect.

Work Experience

S

Scotiabank

Jun. 2026 - Present
AI Engineer
  • 39K+ monthly users
  • 4 agents to production
  • 642 vulnerabilities found
I joined Scotia Intelligence, the bank's central AI platform, as one of its founding engineers. The platform has more than 39,000 monthly active users and over 28 million actions, and I built its conversational search: retrieval-augmented generation that answers with inline citations across the platform's tools, FAQs, and training content. Most of my time now goes to taking model-agnostic agents to production. I lead the deployment of four agentic AI applications on the bank's LangGraph-compatible agent platform, through test, UAT, and production on Kubernetes with Argo CD, Helm, Terraform, and Microsoft Entra ID. I also developed a Word-to-PowerPoint agent commissioned by the bank's Strategy Director, with reusable slide components and four executive slide templates, and shipped four knowledge agents in English, French, and Spanish for procurement, accounting, and application-security questions. Earlier, I led an AI-assisted security scan of 21 applications, 274 repositories, and over 4.3 million lines of code that surfaced 642 vulnerabilities for remediation. I also built a Python analytics and forecasting pipeline over 1.26 million Cloud Run log records, proposed a BigQuery architecture, and mentor an AI engineering intern building a human-in-the-loop RAG agent.
M

Microsoft

May 2025 - Aug. 2025
Software Engineer Intern
  • 2x faster than NCCL
  • Custom CUDA kernels
  • H200 clusters
I spent this internship deep in the GPU stack, building custom CUDA kernels for tensor-parallel LLM inference on NVIDIA H200 clusters. I implemented one-shot, two-shot, and user-buffers All-Reduce variants integrated as TensorRT-LLM plugins in C++, and developed BLAS primitives (GEMM, AXPY, dot product, reduction) benchmarked head-to-head against NCCL and cuBLAS across FP16 and FP32. The highlight was an IPC-based AllReduce that hit 2x speedup over NCCL on sub-2 MB payloads by fusing reduction directly with matrix multiplication, eliminating a full round-trip across the memory hierarchy. I also wrote Triton GPU kernels for broadcasting, activations, and AllReduce, with Python tooling to check each kernel's correctness.
A

Aptum

May 2024 - Apr. 2025
Cloud Developer Intern
  • 66% lower latency
  • 2x CI/CD throughput
  • Monolith to gRPC
I owned cloud infrastructure and service modernization for enterprise clients running workloads across AWS, Azure, and GCP. On the infrastructure side, I managed production Kubernetes with Terraform and Helm and doubled Jenkins CI/CD throughput by redesigning the pipeline stages. The bigger challenge was refactoring a legacy monolith into gRPC-based microservices using Spring Boot and MySQL, which cut request latency by 66% and made each service independently deployable. I also wrote the DevOps runbooks and onboarding guides that three intern cohorts after me used to get up to speed.
S

Sigma Foodservice

May 2023 - Aug. 2023
Software Engineer Intern
  • 80% better address accuracy
  • 98.75% less manual QA
I joined Sigma to improve data quality and developer velocity across one of Mexico's largest food distribution networks. My first project was integrating the Copomex geolocation API into C#/.NET services across 10,000+ retail locations, which pushed address-resolution accuracy up by 80%. I then built out full end-to-end regression coverage using Playwright and Jenkins, cutting manual QA effort by 98.75% and giving the team the confidence to ship faster. Throughout the term I worked closely with product stakeholders in Agile sprint reviews to keep engineering scope and delivery timelines aligned.

Skills

CC++C#JavaJavaScriptTypeScriptPythonGoRustDartSQLSolidityAzureAWSGCPKubernetesDockerHelmArgo CDTerraformOpenTelemetryKafkaRabbitMQMySQLPostgreSQL
RedisMongoDB.NETSpring BootExpressDjangoFastAPIReactReduxFluttergRPCPyTorchTensorFlowKerasNumPyCUDAGeminiLangChainLangGraphTritonHugging FaceOpenAIClaudeJenkinsGitHub Actions
My Projects

Check out my latest work

I've worked on a variety of projects, from simple websites to complex web applications. Here are a few of my favorites.

Headnote

Headnote

A document-grounded chat app: FastAPI and LangGraph retrieve passages from Chroma, and answers cite the source documents in a React UI.

Python
FastAPI
LangGraph
ChromaDB
SQLAlchemy
React
TypeScript
Vite
Docker
LangSmith
Signposted

Signposted

Trip planner where a four-step LLM pipeline (research, itinerary, budget, review) builds a plan behind a FastAPI API, shown on a departures-board web UI.

Python
FastAPI
LangGraph
Groq
PostgreSQL
SQLAlchemy
React
TypeScript
Vite
Docker
charlm

charlm

Character-level name generators, from a count bigram to a BatchNorm MLP and a hierarchical tree, packaged as a typed Python library with a CLI.

Python
PyTorch
pytest
mypy
ruff
GitHub Actions
semantic-release
gpt-from-scratch

gpt-from-scratch

A GPT-2-architecture transformer in PyTorch with a character-level training CLI on Tiny Shakespeare, built following two Karpathy lectures.

Python
PyTorch
pytest
mypy
ruff
GitHub Actions
semantic-release
BPE Tokenizer

BPE Tokenizer

A byte pair encoding tokenizer in Python with a GPT-4-style regex split, special tokens, JSON save/load and a small CLI.

Python
regex
pytest
mypy
ruff
GitHub Actions
semantic-release
autograd-from-scratch

autograd-from-scratch

A scalar reverse-mode autograd engine and small MLP library in pure Python, built while following Karpathy's micrograd lecture.

Python
pytest
mypy
ruff
GitHub Actions
semantic-release
Graphviz DOT
Deep Learning with Keras

Deep Learning with Keras

Eight Colab notebooks that move from NumPy forward and backward propagation to Keras regression, MNIST, CNNs, attention and pretrained models.

Python
Jupyter
Google Colab
NumPy
Keras
TensorFlow
VGG16
BLIP
PyTorch
Multi-Agent Research Assistant

Multi-Agent Research Assistant

A local-first research app that runs a Search, Reader, Writer and Critic pipeline and streams each stage to a React UI.

Python
FastAPI
LangChain
Gemini
Tavily
React
TypeScript
Vite
Playwright
Personal Website

Personal Website

Personal portfolio site on Next.js 14 with a typed content file, an MDX blog and a PR-to-release workflow.

Next.js 14
TypeScript
Tailwind CSS
shadcn/ui
MDX
framer-motion
GitHub Actions
semantic-release
ChiselWare IP Factory Platform by Rocksavage Technology, Inc. (Capstone)

ChiselWare IP Factory Platform by Rocksavage Technology, Inc. (Capstone)

A full-stack semiconductor IP marketplace co-built with a 6-engineer team for verified submission, certification, and sale of Chisel-based IP cores, with Stripe checkout, LinkedIn OIDC, a Chisel-to-Verilog compilation worker, and deployment to Azure Kubernetes Service with Terraform.

React
TypeScript
FastAPI
Python
GitHub Actions
MySQL
Docker
Kubernetes
Azure
Terraform
Stripe
Playwright
Cross-Chain Rebase Token

Cross-Chain Rebase Token

Foundry/Solidity project: an ETH vault that mints a rebase token with per-user interest rates, and a Chainlink CCIP token pool, tested on Sepolia and Arbitrum Sepolia forks.

Solidity
Foundry
OpenZeppelin
Chainlink CCIP
GitHub Actions
semantic-release
DeFi Stablecoin

DeFi Stablecoin

A decentralized stablecoin protocol backed by exogenous collateral (ETH and BTC), featuring algorithmic minting and a USD peg, inspired by the MakerDAO model.

Foundry
Solidity
OpenZeppelin
GitHub Actions
Ride Sharing

Ride Sharing

Go microservices backend for a ride-hailing flow: route and fare preview, trip creation, driver matching over RabbitMQ, and Stripe Checkout sessions.

Go
gRPC
Protocol Buffers
RabbitMQ
MongoDB
Stripe
OpenTelemetry
Jaeger
Kubernetes
Tilt
Docker
WebSocket
NFT Collection

NFT Collection

A collection of ERC-721 NFT smart contracts built with OpenZeppelin, supporting both static and dynamic on-chain traits with fully verifiable metadata.

Foundry
Solidity
OpenZeppelin
Github Actions
ERC20 Token

ERC20 Token

An ERC-20 token smart contract built with OpenZeppelin, deploying Axel Token (AXL) with full test coverage and an automated deployment pipeline.

Foundry
Solidity
OpenZeppelin
Github Actions
LootLark

LootLark

Full-stack video game store on ASP.NET Core, React and .NET Aspire, with Stripe test-mode checkout and a queue-backed worker that issues game codes.

.NET 8
ASP.NET Core
React
TypeScript
PostgreSQL
.NET Aspire
Stripe
Azure Service Bus
Keycloak
Testcontainers
Deep Learning Fundamentals, CNN, Transfer Learning, and ViT Replication

Deep Learning Fundamentals, CNN, Transfer Learning, and ViT Replication

A hands-on collection of PyTorch deep learning workflows covering fundamentals, image classification with CNNs, transfer learning with experiment tracking, and a paper replication of the Vision Transformer (ViT).

Python
PyTorch
CNN
Transformers
Matplotlib
NumPy
ChoreShare

ChoreShare

A cross-platform app for roommates to organize and assign household chores together; built on a Spring Boot microservice backend with a React/Redux frontend and deployed on GCP Kubernetes.

GCP
Terraform
Kubernetes
Docker
Spring Boot
Java
Expo
React
Redux
TypeScript
JavaScript
Jenkins
PostgreSQL
Netlify
MyNotes

MyNotes

A minimal, offline-first note-taking mobile app built with Flutter and Dart, designed for capturing ideas and reminders with local SQLite storage and Firebase sync.

Flutter
Dart
Firebase
SQLite
Github Actions
Game Heaven

Game Heaven

A full-stack e-commerce storefront for video games, featuring an AI customer support chatbot that automatically handles support tickets using the latest GPT models.

Vite
React
TypeScript
TailwindCSS
Express
Node.js
Javascript
MongoDB
Github Actions
Stripe
Netlify
Railway
The Last Show

The Last Show

A serverless web app that generates personalized AI-written obituaries from user input; powered by OpenAI for text generation, AWS Polly for speech synthesis, and Cloudinary for media storage.

Terraform
React
Javascript
Cloudinary
Python
AWS DynamoDB
AWS Lambda
AWS Polly
AWS Secrets Manager
OpenAI
Netlify
Lotion Plus

Lotion Plus

A serverless notes app deployed on AWS Lambda and DynamoDB with infrastructure managed via Terraform and automated deployments through GitHub Actions.

React
JavaScript
Python
AWS Lambda
AWS DynamoDB
Terraform
GitHub Actions
Netlify
LibCode

LibCode

A Python library of data structures and algorithms published to PyPI, with comprehensive unit tests and an automated release pipeline through GitHub Actions.

Python
PyTest
PyPi
GitHub Actions
Certifications

Built on a Foundation of Learning

Learning never stops. I’ve earned 2 certifications across cloud and systems technologies to stay sharp and grounded in the platforms I build on, ensuring every solution I deliver reflects current best practices.

Contact

Let's Work Together

Whether you have a project in mind, a role you think I'd be a great fit for, or just want to say hello, I'd love to hear from you. Drop me an email and I'll get back to you as soon as possible.