Open to backend & AI roles
Backend & AI Engineer
I design RAG architectures and orchestrate LLMs in production — from vector retrieval pipelines to the cloud infrastructure that serves them. Lately that also means writing flight software for a satellite the size of a soda can.
Backend, cloud and applied AI across health systems, infrastructure and CAD automation.
Remote
Remote
Retrieval pipelines, on-device inference and the infrastructure underneath. Every one of these went from empty repo to something someone could actually open.
Google AI Partner Catalyst · Jan 2026
A voice-first legal assistant over Peruvian legislation. Users ask a question out loud, the system retrieves the governing statutes and explains them in plain language instead of legalese.
Dynamic context injection keeps answers anchored to retrieved statutes, cutting hallucinations on legal questions.
Google Chrome Built-in AI Challenge · Nov 2025
A learning platform that runs Gemini Nano directly inside Chrome. Flashcards, summaries and exercises are generated from your PDFs without a single byte leaving the browser.
100% offline, privacy-first inference on-device — with a FastAPI + LangChain path for the heavier cloud work.
NASA Space Apps Challenge · Oct 2025
A knowledge engine for space-biology literature. It clusters hundreds of papers by meaning, labels each cluster automatically and lets you walk the result as an interactive map.
SciBERT embeddings + HDBSCAN clustering, with an AI agent that extracts research evidence from each paper.
Open source · maintained
The authentication boilerplate I kept rewriting, packaged once and properly: Google OAuth, JWT in HttpOnly cookies, Prisma 7 and a decoupled backend, all up with one Docker Compose command.
Tokens live in HttpOnly cookies rather than localStorage, and every env var is validated with Joi at boot.
Public health · Android release
A mobile app for anemia screening that corrects haemoglobin readings for altitude and anthropometry — the adjustment that decides whether a diagnosis in the Andes is right or wrong.
Built with regional health staff as the end users, and shipped as a signed Android release.
UNSA · Research project, Jun — Dec 2025
Machine learning models that classify crops and detect phenological stages from satellite imagery, built for agricultural monitoring across the Arequipa region.
+75% classification accuracy, and an OpenCV preprocessing pipeline that cut manual analysis time by 80%.
Thirty-odd more repositories live on GitHub .
ONISAT CanSat 2026 is a full mission stack I work on end to end: firmware on the flight computer, a LoRa downlink, and the ground software that turns raw packets back into telemetry and stereo images.
The part I'm proudest of: radio packets get lost mid-flight, so the ground station reconstructs the missing image regions with OpenCV inpainting instead of dropping the frame.
Signal chain
Onboard
Sensor array
Altimeter · IMU · magnetometer · environmental · power
Flight software
C++17 on Raspberry Pi Zero 2W
Mission FSM · stereo capture · servo control
Radio link
LoRa SX1278 → ESP32-S3
5 Hz telemetry with interleaved image fragments
Ground backend
Python · OpenCV · WebSockets
Packet reassembly · inpainting recovery
Dashboard
Next.js 16 · React 19 · Three.js
Live 3D attitude · anaglyph stereo imagery
Around the mission
Python and TypeScript day to day, retrieval pipelines on top, Docker and one of the three big clouds underneath.
Highlighted items are what I reach for most weeks — the rest I've shipped with, but less often.
OpenAI API · FAISS · Vector DBs · Agentic workflows
Node.js · Java · Spring Boot · REST · WebSockets
Azure · Terraform · GitHub Actions · Grafana · Linux
PostGIS · MongoDB · Prisma · OpenCV · NumPy
Astro · Angular · React Native · Tailwind CSS
LoRa / SX1278 · Arduino · I2C sensors
Where the work was judged by someone other than me.
Degree
Universidad Nacional de San Agustín de Arequipa
Arequipa, Peru · Expected 2027
Alongside the degree I run research and competition work through the university: satellite imagery ML for crop monitoring, and the ONISAT CanSat mission.
Languages
Also worth knowing
I work remotely from UTC−5, which overlaps a full working day with both American coasts and a morning with Europe.
Backend, applied AI or infrastructure — if you're hiring for any of those, or just want to compare notes on RAG systems, my inbox is open. I reply within a day.