Software Engineer · Backend, Full-Stack & AI
Matteo Pesenti
Software engineering student at the University of Waterloo. I work across the whole stack, and everything I build these days is AI-first.
About
I’m a software engineer who works across the whole stack, with a pull toward backend systems and, lately, the AI infrastructure around them: on-call agents, MCP servers, and the tooling that keeps them dependable in production.
I graduate in April 2027 and I’m looking for full-time backend, full-stack, and AI engineering roles in the US.
Outside of that, I spend most of my free time mountain biking, playing poker, and shooting pool.
Experience
- Worked directly with customers to build human risk scoring, a Django + React model that flags where a company's risk is coming from, unblocking 9 enterprise deals worth $2M in ARR.
- Architected an internal MCP platform in Django + Terraform that gives AI agents PII-masked, audited access to production data, Airflow, and Kubernetes to diagnose failures, cutting failed pipeline runs from 12% to 0%.
- Engineered an AI agent that autonomously resolves production incidents, restarting failed DAGs, opening fix PRs, and suppressing noisy alerts, cutting manual intervention by 65%.
- Set up cloud dev environments that give engineers and coding agents a running copy of the app to work in, so agents can build and test changes in parallel, driving 30% more PRs.
- Built an LLM extraction pipeline that parses option grants and stock certificates from uploaded documents into structured cap table data, writing the prompts and extraction code, handling 1,400 extractions/day.
- Redesigned cap table extraction as a fully async experience in Django + Celery, tracking job status live and emailing users a deep link when results are ready, removing a 2-10 minute blocking wait.
- Designed Braintrust eval scorers that grade every prompt change against 100+ labeled documents, lifting extraction accuracy from 80% to 99.6%.
- Rebuilt the cap table importer and warrant-signing pages as React + TypeScript microfrontends with module federation, serving 12,000 weekly users.
- Implemented a customer portal editor in React + Go, building the UI and the REST and gRPC APIs that let businesses customize their WiFi portals, as part of a $5M ARR deal with Verizon.
- Integrated Foursquare, Google Places, and Yelp APIs with rate-limited, retrying clients to enrich venue metadata powering audience segmentation across millions of WiFi users.
- Built a React component for dynamic captive-portal ads backed by gRPC services querying PostgreSQL and S3, enabling targeted campaigns.
- Created a Go testing framework for PostgreSQL with schema cloning and parallel CI execution.
- Developed a React + Flask wildfire hazard-mapping tool, building the map UI and serving geospatial data through REST APIs, cutting hazard assessment from days to minutes.
- Developed a Python + Qt desktop tool to disseminate wildfire-resilient building guidelines, integrating hazard, exposure, and construction GIS data.
- Built diagram solving in FastAPI + React for photographed homework problems, fine-tuning a PyTorch classifier on 10,000 labeled images to detect diagrams, lifting engagement 8% in A/B tests.
- Migrated ML services from Hugging Face to Baseten, reducing latency spikes by 30% across 20+ production models.
Contact
Have a role, a project, or just want to talk shop? Send a message below, or email me directly