Lima, Peru · Data & AI Analyst at Izipay
I'm Mauricio Alvarez. I build AI agents and automations that run in production, and I research interpretable computer vision, reinforcement learning and models that run on the device.
01
Four recent pieces of work, from research to product. All the code is public.
Research into how efficient vision transformers organise the concepts they learn. It builds on the Visual Concept Connectome (CVPR 2024) and takes it to the SHViT family.
A native SwiftUI app that runs language models entirely on the phone. No account, no backend and no telemetry: weights are downloaded and stay on the device.
Reproduces OpenAI's multi-agent hide-and-seek environments in Docker, trains agents with PPO and self-play, and transfers those policies to a cyber benchmark built from real vulnerabilities.
A platform that takes invoice PDFs and extracts their data automatically with language models and Document AI. Deployed on Google Cloud, with sign-up and a dashboard.
YOLOv8 detection with Kalman-filter tracking. Zone-based counting registered 80 vehicles on a segment where line crossing only saw 52.
Upload a portrait and get five studio looks through Stable Diffusion img2img. Runs on a local GPU or through an API, with a face lock to preserve identity.
An Angular 19 app that queries Fintual's public API and charts a fund's evolution over the date range you choose.
Sequential file and extendible hashing implemented from scratch, with a SQL parser and a GUI to run queries and measure timings.
An inverted index built with SPIMI, ranked by TF-IDF and cosine similarity over arXiv papers, and compared against PostgreSQL's index.
KNN and range search over face embeddings, comparing a sequential scan with R-tree and KD-tree indexes.
02
I care about understanding why a model decides what it decides, and about building what it takes for that answer to reach someone: the training, the API, the interface and the deployment.
I graduated in Computer Science from UTEC, in the top fifth of the programme, and today I work on Izipay's Data & Analytics team. My personal projects range from file structures written in C++ to native iOS apps and cloud services.
03
The summary is here; the full detail is in the PDF.
Jul 2026 – Present
Izipay · Lima, Peru
Joined the Data & Analytics team after the Summer Programme. I maintain and evolve the AI-agent solution that automates the post-sales email channel, and support MLOps pipelines with LangChain, LangGraph, LangFuse and LangSmith.
Jun 2026 – Jul 2026
LineLogic AI · Remote
Contacted after an ElevenLabs hackathon to build a voice-agent flow for customer service. I developed the agent logic and its tools in Python, the company website, and a dashboard to monitor conversations and agent performance.
Jan 2026 – Jul 2026
Izipay · Lima, Peru
Selected from more than 4,000 applicants. I built the automation of the email support channel with agents in Python and LangGraph, for an estimated 10% saving in total email-processing time.
Feb 2025 – Aug 2025
Crece Capital Factoring · Lima, Peru
Automated PDF validations with Python and AI, cutting processing per file from 15 minutes to 1. Developed and maintained web solutions in AngularJS, C# and TypeScript.
2021 – 2026
UTEC — Universidad de Ingeniería y Tecnología
Top fifth of the programme.
2019 – 2020
International Baccalaureate
04 · Contact
I'm open to opportunities and collaborations in AI and software development. Write to me and let's talk.