Engineering Manager at Platzi · Córdoba, Argentina

JaviGuignard.

I build software, lead engineering teams and have been teaching programming since 2008. Lately I keep coming back to one question: what do we need to remove so AI can actually make us faster?

Read talk excerpts Invite me to speak
Javi speaking on stage, microphone in hand
  • Platzi Conf Bogotá 2026
  • Cloudflare Connect Las Vegas 2025
  • BeerAI Córdoba 2025
  • Platzi Conf Bogotá 2025
  • Docker Bday #6
  • FLISoL Córdoba
  • PyDay UNC
  • Universidad Blas Pascal

01 · Talks

Things I've said on stage

Excerpts from recent talks. I still stand behind most of them, and one I corrected in public a year later.

  • Keynote
  • Platzi Conf 2026
  • Bogotá
  • Aug 2026

Descaling first

What we had to shut down for AI to work

A three-day hackathon where we didn't build a single new feature. We folded four payment microservices back into the monolith, deleted 250,000 lines of dead frontend and moved B2B out to serverless functions. This talk covers what that taught me about architecture, about teams, and about the order in which you change them.

  • 3 days0 new features
  • −250,000lines of obsolete frontend
  • 4microservices shut down
  • 8–10 → 1–3people per initiative, with ownership

Excerpts · Descaling first

01 / 09
  1. 01

    AI doesn't add speed on top of just any system. It gives speed to whoever has the nerve to remove the complexity that no longer buys anything.

    Descaling firstmin 00:00

  2. 02

    Productivity wasn't how many things we added; it was how much friction we stopped paying on everything that came after.

    Descaling firstmin 01:35

  3. 03

    At Platzi, almost half of our pull requests are no longer opened directly by a person.

    Descaling firstmin 04:50

  4. 04

    Your architecture is a UX decision. And a huge share of your users are now agents.

    Descaling firstmin 05:25

  5. 05

    I'm not proposing a return to the monolith as a religion. The rule is to stop paying for separations that don't buy anything.

    Descaling firstmin 10:20

  6. 06

    If an agent produces ten times more changes and our ability to understand, test and observe doesn't change, we don't have ten times the productivity: we have ten times the risk waiting at a gate.

    Descaling firstmin 13:25

  7. 07

    As long as the org chart stays the same, the architecture fills back up.

    Descaling firstmin 17:20

  8. 08

    The skill gaining the most value is directing work you didn't do yourself.

    Descaling firstmin 22:45

  9. 09

    The next wave of engineering leaders won't be the ones who scaled best. It'll be the ones who had the nerve to descale first.

    Descaling firstmin 23:35

Correction

2025

Start with the technical layer. Simplify the code and coordination and the organization will follow on their own.

2026

At Platzi it happened the other way around. First we changed the organization, and only then could we look at the architecture without defending every service as some team's territory.

It was an elegant story. It was also incomplete.

Descaling first, min 16:10
  • Breakout session
  • Cloudflare Connect 2025
  • Las Vegas
  • Oct 2025

From Algolia to edge-first semantic search

AI has democratized world-class search. What once cost millions to build is now within reach.

How Platzi replaced Algolia with its own semantic search on Cloudflare Workers, Vectorize and AI Gateway. Search, instant search and recommendations share one relevance layer that re-ranks results for each student. Then the inverse recommender: you go in with a course and come out with the students most likely to want it, using Hyperdrive and pgvector once Vectorize's 20-result limit got in the way. Targeted campaigns converted 5x better.

With Edwar Barón and Juan José Torres

  • Workers
  • Vectorize
  • AI Gateway
  • KV
  • Hyperdrive
  • pgvector
Session page
  • Ignite, 5 min
  • Platzi Conf 2025
  • Bogotá
  • 2025

We gained 4 months of life thanks to a mindset shift reinforced by AI

Five minutes on how using AI to strip operational friction out of our workflows gave the team months back. I argued you should start with the technical layer. A year later I went back on stage and said I'd got the order wrong.

  • Productivity
  • AI in engineering
  • Community talk
  • BeerAI
  • Filadd HQ, Córdoba
  • 2025

Specs that drive AI development: clear inputs, stronger outcomes

Vibe coding versus specs, live. I started with a one-shot prompt, then wrote a spec and let Claude Code and a subagent build a ChatGPT clone (ChamuGPT) on top of our AI gateway, deployed to Cloudflare Workers. A structured spec lines up requirements, evaluation criteria and the feedback loop between engineering and business.

  • Spec-driven development
  • Claude Code
  • Agents

Earlier talks

  1. 2019Docker for babiesDocker Bday #6, Córdoba
  2. 2018Git: what it is and how to use itUniversidad Blas Pascal, Córdoba
  3. 2017Building my own study environments with Docker and docker-composeUniversidad Blas Pascal, Córdoba
  4. 2015Django, from zero to a real appFLISoL Córdoba
  5. 2012Python in high school: using Python to teach CSPyDay, UNC

02 · About

Engineer, manager, and a teacher the whole way through

Black and white portrait of Javi at a lectern, looking up

I'm an electronic engineer from UTN Córdoba and I started coding at 15. My first job, in 2004, was industrial software: SCADA, HMIs, PLCs. After that came TCP servers in C and Java, a software company I co-founded and ran for six years, the backend for Argentina's national weather radar website, and three years as tech lead at Claritywave running Django on AWS Lambda.

I joined Platzi in 2021 and I'm an engineering manager there today. My work sits where AI, education and enterprise software overlap, including AI systems that turn what a company knows into learning experiences that scale. I care about small teams with real ownership, people who understand the business problem behind the ticket, and having the nerve to delete what no longer pays for itself.

Since 2008 I've taught programming at ITS Villada, the same technical school I graduated from. Python, Java, Django and databases for years; now a course on new technologies where students work with AI, cloud, Git and containers, the same tools I use at work. From 2020 to 2022 I also taught Scrum, Git, Docker and web services at Universidad Blas Pascal.

Good solutions start with a conversation about the customer's real problem.

  • 15my age when I wrote my first program
  • 2004first job writing software
  • 2008teaching programming ever since

Path

  1. 2021 – nowEngineering Manager, Tech LeadPlatzi
  2. 2018 – 2021Sr. Software Developer, Tech LeadClaritywave
  3. 2017 – 2019Full stack developer and technical PM, national weather radar websiteGrupo Radar Córdoba
  4. 2016 – 2018Django consultantEmser S.A.
  5. 2011 – 2021IT consultant and freelance developerOmixom Ingeniería
  6. 2010 – 2016Co-founder, full stack developerOsninet Sistemas y Redes
  7. 2008 – 2010Backend developer (C for PIC, Java, Python)Omixom Ingeniería
  8. 2004 – 2008Industrial software: SCADA, HMI, PLCR.C. Ingeniería

Teaching

  1. 2008 – nowProgramming teacherITS Villada
  2. 2020 – 2022University professor, Informatics Lab IVUniversidad Blas Pascal

Education

  1. 2016Electronic EngineerUTN, Facultad Regional Córdoba
  2. 2002Industrial Electronics TechnicianITS Villada

03 · Topics

What I can talk about

  1. 01

    Descaling for AI

    Which services, ceremonies and roles to retire so AI speeds you up instead of becoming one more layer on top.

  2. 02

    AI search and recommendations at the edge

    Embeddings, Vectorize, per-user re-ranking, and shipping it safely with feature flags, canaries and A/B tests.

  3. 03

    Spec-driven development with agents

    Specs an agent can actually execute, and evaluation criteria that tell you whether the result is any good.

  4. 04

    The product engineer

    Small teams, wide ownership, and individual contributors who now need manager skills to direct a team of agents.

  5. 05

    Teaching tech when AI writes code

    How I bring AI, cloud, Git and containers into a high school programming class.

Keynote, technical session, workshop or 5-minute ignite. In Spanish or English.

04 · Contact

Got an event, a team or a tangled architecture?

Send me a message on LinkedIn. I'm @kaajavi on X and GitHub.

Message me on LinkedIn X · @kaajavi GitHub · @kaajavi