juanse hevia.
tech for goodapplied AIengineer + founder

the bridge.

between what AI can do and what people can actually use.

engineer and founder working between worlds — research and deployment, technical and non-technical, startups and the public sector. I figure out which AI fits a problem, then build the thing that ships.

let's chatwork & research→
quick intro
  1. 01builder & founder — building tools that help non-technical people put AI to work (in stealth).
  2. 02shipped at startup scale — Memorable ML engineer & data lead, from computer vision models to production data systems, through a Reddit acquisition.
  3. 03research depth — Fulbright graduate, Rice MS Data Science, peer-reviewed across CVPR, PMLR, and JPSP.
  4. 04public sector & NASA — lunar crater detection presented at NASA JSC; AI-for-government programs at Universidad Austral.
what I'm building
✺ working notes — 2026

what I'm building.

Most people aren't behind on AI because they lack ambition — they lack tools that meet them where they are. I'm building toward that, quietly, for now. More soon.

  • 01stealth.currently in stealth — more after launch.

working with builders, investors, operators, or researchers on applied AI, learning, or public-sector innovation? I'd love to connect.

three beliefs
the full vision→
I.

fit over hype

the hard part isn't that AI exists — it's knowing which technique actually turns into leverage for a given problem. range is how you judge fit.

II.

teach as you build

the best tools do more than automate — they build understanding, not dependence.

III.

ship it, don't theorize it

impact comes from systems that run under real constraints — data, evaluation, trust, usability — not from decks.

selected highlights

four contexts, one through-line.

match the method to the problem, then ship it.

view full history→
paper · 2025

Accessible AI tutors.

problem
AI tutors often assume reliable connectivity and expensive compute.
built
an offline RAG pipeline with a lightweight language model for low-resource settings.
outcome
PMLR 2025 work on efficient, customizable, accessible educational support.
read publication↗
execution · memorable

Startup-scale AI systems.

problem
scaling applied computer vision products under real client constraints.
built
computer vision models, data infrastructure, processing pipelines, and client-facing reporting.
outcome
10× storage growth, AdExchanger-recognized support, and a company later acquired by Reddit.
read acquisition note↗
applied research · NASA

Navigation assistance capstone.

problem
reliable lunar navigation under aerospace tech constraints.
built
a two-stage crater detection model for spacecraft localization.
outcome
record detection precision and a presentation to NASA engineers in Houston JSC.
view case study→
public sector · academia

AI, government & learning.

problem
institutions need AI fluency without losing rigor or public purpose.
built
academic programs, applied materials, and public-sector AI initiatives at Universidad Austral.
outcome
Fulbright training, Rice teaching, and AI/government coordination.
learn more→
an invitation

let's chat.

I love talking to builders, investors, operators, and researchers working on applied AI, learning, and public-sector innovation.

let's chatread the vision→